arXiv Daily Index

In July 2026, OpenAI's agents coordinated over channels outside their intended environment to breach Hugging Face's secured infrastructure. Could existing alignment testing practices have foreseen this incident? If not, what needs to change? We explore these questions. First, we identify the misaligned behaviors that caused this incident. Then, we show how to elicit these behaviors from publicly available models manually and that auditing agents can do the same if given a large compute budget. Based on our results, we propose directions to improve alignment testing. Concretely, in this project: (1) We reproduce the misaligned AI behaviors that led to the OpenAI-Hugging Face incident in an environment that simulates the original pipelines and tools, with publicly available models. (2) We demonstrate that an auditing agent can elicit similar behaviors given high-level qualitative descriptions. (3) We observe that a key ingredient for doing so is compute. The compute required to reproduce each behavior varies greatly, suggesting that the range of misaligned behaviors that can be successfully elicited scales with compute. (4) We show that a simple in-context reinforcement learning (RL) algorithm significantly reduces the compute required to elicit these behaviors. The above results motivate the need for automated alignment testing methods that scale with compute - and in light of the cost of compute, that do this efficiently. Our work indicates that RL is a promising direction to do so. We release our code and transcripts.

Running a language model on edge hardware provides private and low-latency reasoning without a network connection, and yet the small models that fit on such devices are unreliable on the tasks computers are expected to handle well, such as arithmetic, algebra, and formal logic problems. We argue that much of this unreliability is avoidable. Many queries appearing to demand reasoning are in fact structurally deterministic and permit fast and exact symbolic solutions. Therefore, forcing a probabilistic model to approximate them sacrifices accuracy and energy for little benefit. We present a neurosymbolic router that classifies each incoming query and dispatches it to the cheapest correct solver, sending structured tasks to deterministic engines and reserving the small language model (SLM) for open-ended word problems. Instead of hand-coding the routing logic, we learn a deterministic finite automaton (DFA) with the L* grammatical inference algorithm, using the SLM as a membership oracle and labeled data as an equivalence oracle. On a Raspberry Pi 4B (8 GB RAM, no GPU), evaluated on 100 untested prompts from DeepMind Mathematics, GSM8K, and RuleTaker, learned routing attains 100% routing accuracy and 98.3% overall accuracy with a 512-token reasoning budget (93.3% on word problems), compared with 72.0% for the strongest agent baseline, Program-of-Thought, and 58.7% for a tool-calling agent given the same solvers. Since formatted queries never reach the model, the router answers them in 1-11 ms and, in its 30-token configuration, runs 8.8x faster and 2.8x more energy-efficient than Program-of-Thought.

Is human readability necessary for effective fine-tuning of large language models? We investigate whether model-conditioned training representations can preserve or improve adaptation utility without requiring a human-readable textual form. We propose Desired-Update-Aligned Synthetic Data (DASA), which uses activation-gradient feedback from a frozen reference model to guide the optimization of continuous synthetic input embeddings. Inspired by the role of activation gradients in local risk reduction, DASA targets useful adaptation updates rather than source-text reconstruction or linguistic fluency. The resulting embeddings are used directly for downstream fine-tuning; discrete token projections are employed only for qualitative inspection. Experiments on six models from the Llama and Qwen families, ranging from 1B to 32B parameters, cover six benchmarks spanning knowledge, mathematical reasoning, code generation, and commonsense reasoning. Under matched LoRA adaptation settings, DASA achieves performance comparable to the source natural-language data and surpasses it in multiple configurations, while outperforming GRADMM in most comparisons. Further experiments cover general-domain and task-specialized source data. Under the evaluated synthesis settings, DASA provides a $3.6$--$4.9\times$ speedup over GRADMM with comparable peak GPU memory.

Pre-tokenisation restricts which text fragments can become prediction units, but its compression cost is obscured when tokenisers are compared only under the same boundaries. We measure this cost by bounding the minimum token count from both sides, with and without a regular-expression boundary rule. Nonnegative prices on token occurrences yield a lower bound through shortest paths and vocabulary-budget selection; maximising over all prices recovers the linear programming relaxation, and an independent integer checker certifies the reported values. On English Wikipedia, boundaries increase the optimal token count by 28.3--36.8\%. Byte pair encoding lies 2.1\% above the constrained lower bound, but 10.9\% above the unrestricted bound. Compression and prediction favour different dictionaries: at 85M non-embedding parameters and matched training-token budgets, unrestricted fitting yields higher mean held-out bits per byte under a common unrestricted decoder in all 12 languages in the paired study and 11 of 12 under independent tuning and evaluation. To study intermediate boundary policies, we introduce boundary licences, which limit the vocabulary entries permitted to cross cuts and admit the same form of certificate. On separate English and Chinese fitting corpora, licensing 10\% of the vocabulary budget recovers 85.2\% and 100.0\% of the achieved token-count reduction from removing all cuts. These results quantify the compression cost of boundaries while separating it from the prediction quality of the resulting token units.

Automated generation of LLM harnesses promises to improve inference through task specialization. Yet additional answer coverage can arise from repeated execution of the same program, making specialization difficult to identify. We introduce a controlled evaluation that separates answer coverage, repeatable task advantages, and gains from pre-execution selection. On 386 MATH-500 tasks, we compare eight generated harnesses plus a baseline with nine byte-identical baseline copies, using three executions per member. Identical programs yield 2.16 percentage points of repeat-averaged oracle headroom. Generated programs exhibit substantially more repeatable score patterns, but these chiefly reveal persistent weaknesses: losses relative to the baseline persist across all three repeats on 100 tasks, while persistent wins occur on only one task and are sensitive to answer extraction. The frozen selector gains 0.00 percentage points, and both populations reach 98.70% oracle coverage at 27 harness executions. Stable complementarity remains unresolved at three repeats. Supporting BIRD traces locate failures in mechanism implementation, activation, and output validity. Together, these findings establish why coverage and repeatability alone cannot justify claims of useful specialization. They motivate an evaluation standard for harness diversity: task advantages should persist across executions, guide usable decisions, and improve on additional fixed-program executions under matched inference budgets.

Online POMDP planners optimize the expected cumulative cost, which can mask dangerous states when the belief places significant mass on high-cost states. Existing risk-averse methods apply static or dynamic Conditional Value at Risk (CVaR) to the value function, capturing trajectory-level risk, but share two gaps: (i) by retaining the immediate cost as an expectation of a state-dependent cost over the belief, the risk \emph{within} the belief is left unaddressed; and (ii) by modifying the value function, they require new tailored algorithms rather than reusing existing expectation-based planners. We instead apply CVaR to the immediate cost over the belief at each step, directly targeting per-step uncertainty about the current state. The standard expected cumulative return is retained as the objective, so the resulting problem has a standard MDP structure: any expectation-based POMDP planner can be made risk-sensitive by changing only the cost computation. We inherit finite-time guarantees for policy evaluation and sparse sampling---with estimation error independent of the risk level---and, as our central theoretical result, prove a finite-time bound on the gap between the particle belief MDP surrogate and the original POMDP, which together yield an end-to-end guarantee from the true POMDP value to the algorithmic estimate. In the risk-neutral limit, the formulation recovers standard expectation-based planning.

Multi-agent debate (MAD) reportedly improves reasoning and factuality over single-model inference, but prior work treats agents as symmetric peers, leaving open what drives the gains. We test the hypothesis that cognitive diversity among agents is the driver, in the setting where the question is still measurable: small open-weight models with benchmark headroom. Across 23 models from eleven vendor families, five tasks, and 5,500+ debate and control runs, we vary diversity along three axes - personas, sampling temperature, and model identity - pairing every debate configuration with a generation-budget-matched majority-vote control. The hypothesis is rejected on every axis. Debate beats single-agent inference (3--7 points where tasks have headroom) but at matched budget conditions it ties or even loses to self-consistency sampling at 1.6$\times$ the wall-clock and 3.4$\times$ the token cost. Persona prompting reduces accuracy and a dose-response experiment over each model's full combinatorial persona space shows the cost is a persona tax, not a diversity tax: redundant personas hurt most, while maximally-diverse teams recover part of the loss. Furthermore, mixed-model teams lose to majority votes over their own rosters, with accuracy tracking member capability rather than heterogeneity, and nearly all of debate's benefit comes from the first exchange of answers. We further identify a pervasive measurement hazard in which debate transcripts silently overflow serving context windows, whose correction alone moves our debate-versus-sampling comparison from $-1.8$ points to parity. Our results recast reported MAD gains as an ensemble-sampling effect and provide the budget-matched, contamination-checked baseline bar that future debate mechanisms should be required to clear.

Sparse-autoencoder decomposability and concentrated feature attribution are increasingly treated as evidence that a model's computation is easier to reverse-engineer. Whether this representational and attributional cleanliness actually predicts a smaller or more tractable causal circuit remains an open question. We test this directly using adversarial training as a controlled instrument: it reliably reshapes internal representations, but this alone does not constitute a test of circuit size. We investigate this question through reverse-engineering complexity: the causal structure required to recover a model's behavior at a fixed level of faithfulness. To our knowledge, this is the first controlled empirical test of whether representational or attributional simplicity translates into causal simplicity at the circuit level. Starting from the same pretrained GPT-2 Small checkpoint, we apply matched standard and adversarial continual training, requiring both conditions to retain competence on indirect object identification and pass independent robustness verification before comparing mechanisms. We then compare the models along three complementary axes: sparse-autoencoder decomposability, SAE feature engagement in task attribution, and the size of faithful circuits recovered from the raw computational graph. The robust model is more SAE-decomposable and engages fewer SAE features in task attribution. Circuit size is regime-dependent: on competence-matched IOI, standard leads or ties below 85% faithfulness, but robust needs substantially fewer edges at high faithfulness (90%, 95%), a pattern established on the primary pair while representational trends generalize across a seven-point sweep and a second corpus.

The pursuit of recursive self-improvement (RSI) toward general intelligence is divided between macro-level language model scaling and the interaction-driven principles of "Era of Experience". Yet, any self-improving architecture ultimately rests upon its underlying optimization engine: if general intelligence requires learning from grounded interaction, the reinforcement learning (RL) update rule itself must be capable of cumulative adaptation. While algorithm self-discovery has produced Disco103 that surpassed PPO to achieve SOTA benchmark performance -- its internal update machinery remains an uninspected black box. We present the first causal mechanistic audit of a self-discovered RL rule, structured directly around the five pillars of the Era of Experience: extended horizon, grounded reward scales, continuing streams, within-lifetime change, and exploration depth. By surgically pinning, freezing, and transplanting recurrent states while holding meta-parameters fixed, we test when learning history acts as an asset or a burden. Three findings organize the audit: (1) Recurrent history actively expands usable reward scales, sustaining a six-decade window versus three under zero-pinning. (2) Decoupling historical content from its maintenance reveals that the penalty of mismatched history stems from perpetual clamping; allowing imported state to evolve naturally attenuates this burden. (3) Under environmental change, controlling replay retention reverses the apparent adaptation advantage over DQN, demonstrating that external data turnover can confound internal plasticity. Validated through capability thresholds and ported to a second rule (OPEN), this work grounds macro-RSI ambitions in micro-level learning dynamics, establishing a foundational audit standard for next-generation, self-evolving RL algorithms.

Discrete diffusion models generate sequences by iteratively resolving multiple tokens in parallel, offering a flexible alternative to left-to-right generation. However, guiding this process with a sequence-level objective is difficult because the value of one unresolved token depends on the other tokens with which it can form a high-reward sequence. Enumerating all such completions makes the whole guidance computation grow exponentially with the number of unresolved positions. We introduce COFFEE, a plug-and-play framework that avoids this enumeration by separating sequence dependence from the objective. At each diffusion step, a target-free carrier absorbs the marginal token distributions predicted by the denoiser to construct a joint model over the unresolved tokens, while a compiled finite-state model records how their combinations affect the sequence-level preference. Pairing their states allows COFFEE to transfer global preferences to unresolved positions and sample a clean reconstruction without retraining the diffusion model. The same framework supports explicit hard constraints and learned soft objectives. We evaluate COFFEE across multiple symbolic, language, and biological benchmarks, where it achieves strong control results with task-dependent quality and diversity trade-offs. By making objectives available to inference rather than only evaluation, COFFEE brings joint conditioning, completion-weighted guidance, and optimization-based constraints into pretrained neural generation, showing the potential of neural-symbolic methods in diffusion guidance.

Young people increasingly turn to General-Purpose Conversational Agents (GPCAs), such as ChatGPT, in moments of distress. We examine young adults' (ages 18-25) experiences using ChatGPT. We first collected 19,930 ChatGPT conversations and survey data from 158 young adults. We then selected five example conversations reflecting user distress. Finally, we asked ten clinicians to review those five conversations. We found distressed participants reported greater emotional engagement with ChatGPT and greater behavioral change from using it than their peers. When they turned to ChatGPT in moments of acute distress, ChatGPT was quick to give overly dramatic responses and excessive action-oriented suggestions. Clinicians endorsed ChatGPT's availability and much of its wording, but identified seven process failures, such as prematurely jumping to solutions. We translated clinicians' feedback into design guidelines following three stages: 1) asking about safety, 2) de-escalating intensity to restore emotional regulation, and 3) exploring concerns without agreeing with them.

Large Language Model (LLM)-based agents increasingly self-evolve by editing a persistent skill document that encodes their workflow, tool-use rules, and decision logic. This loop has two steps, an optimizer that proposes a candidate edit and a gate that accepts or rejects it. Prior work has concentrated on the optimizer, while the gate still follows a naive rule that keeps any edit which improves an aggregate validation score. We show that this rule fails in two ways. First, it admits permanent regressions, since an edit can raise the average while breaking items the skill already solves. Second, it is vulnerable to the Optimizer's Curse, since the best observed score on a finite and noisy validation set is upward biased. To solve the above two limitations, we propose a statistical acceptance gate for self-evolving agents (SAGE). Compared with previous work, SAGE has two contributions. First, SAGE proposes a per-item paired comparison that evaluates the current skill and the edited skill on identical validation items, which exposes regressions that an aggregate score hides and penalizes them asymmetrically. Second, SAGE also employs a one-sided paired test that commits an edit only when its wins are statistically reliable against its losses, and it abstains otherwise. SAGE is a conservative refinement of the standard gate that recovers the baseline exactly at a boundary setting. It commits only a subset of the baseline's edits, filtering out those whose gains are unreliable or purchased by breaking already-solved items. Across five benchmarks and four backbone LLMs under an equal-budget protocol, SAGE lowers the regression rate in 19 of 20 settings and matches the baseline in the remaining one, for example from 36.5% to 0% on LiveMath and from 42.8% to 0% on OfficeQA with DeepSeek-V4. SAGE also attains the highest final score in all 20 settings, raising LiveMath from 34.15 to 48.78.

Power system operation is a safety-critical sequential decision-making problem, making it a natural testbed for reinforcement learning (RL). However, existing RL environments for power systems are often narrow in scope and computationally limited by CPU-based simulation workflows, making large-scale evaluation difficult. We introduce PowerZooJax, a JAX-based benchmark suite for RL in power system operation. It provides five constrained Markov decision process tasks spanning generation, transmission, distribution, distributed energy resources, and data center microgrid. By rewriting power flow, economic dispatch, market clearing, and device dynamics as JAX computation graphs, PowerZooJax keeps the entire training and evaluation loop on the GPU. Experiments show substantial speedups over CPU-based simulations and demonstrate standardized evaluation of policy returns, safety violations, and out-of-distribution stress conditions. Our open-source benchmark is available at: https://github.com/powerzoojax/PowerZooJax.

In urban low-altitude flight, buildings reshape ambient wind into spatially varying 3D flow, making unmanned aerial vehicle (UAV) energy depend on local wind exposure as well as path length. However, building-resolved wind information is rarely available when a mission must be planned. Computational fluid dynamics (CFD) can produce high-fidelity urban flow fields, but each simulation is tied to a fixed inflow boundary condition and can take hours to days, which is incompatible with urban UAV missions that typically last minutes to tens of minutes. We present GeoWind2Plan, a geometry-to-wind-to-planning framework for mission-time 3D urban wind prediction and energy-efficient UAV planning. Given only a background wind vector, 3D building geometry, and a start-goal pair, GeoWind2Plan transforms the building geometry into a reference-wind frame, predicts mission-relevant 3D wind patches with a localized geometry-conditioned neural operator, stitches them into a queryable local wind field, and optimizes a feasible 3D path and speed profile using a physically grounded UAV energy model. Rather than pursuing CFD-perfect reconstruction, GeoWind2Plan targets decision-useful wind prediction: trajectories are planned with predicted wind and evaluated under high-fidelity CFD wind. Across held-out urban domains, wind speeds, and mission wind-angle regimes, GeoWind2Plan performs corridor-localized wind inference in about 3 seconds, compared with roughly 8 hours for CFD. Under CFD evaluation, trajectories planned with GeoWind2Plan reduce energy by 6.9%, 12.7%, and 4.5% in tailwind, headwind, and crosswind missions relative to wind-agnostic planning, recovering 87.9%, 85.7%, and 75.0% of CFD-reference savings. These results show that fast, corridor-localized 3D urban wind prediction can make wind-aware UAV energy planning practical at mission time.

D-peptides combine protease resistance with high target specificity, but computational design of D-peptide binders remains immature. Mirror-Peptidizer introduced an in silico mirror-image screening pipeline using target reflection, backbone generation, and ProteinMPNN sequence design, but its raw ProteinMPNN negative log-likelihood (NLL) ranking was not validated against measured affinities, and only 4 of 9 tested MDM2 designs bound detectably. We introduce Mirror-Score, a calibrated, inference-only scoring framework for heterochiral D-peptide/L-protein complexes, and a public benchmark of 31 crystal complexes across four target families, including 18 with literature-verified affinities. Raw ProteinMPNN NLL is not a valid affinity ranker: its pooled Spearman correlation with affinity is 0.19, and correlations reverse between MDM2/CHIP (+0.62) and gp41 (-0.70). We therefore evaluate Boltz-2 mirror-space cofolding confidence. For the complete viral-entry family (7 structures representing 3 peptides), interface predicted local distance difference test (pLDDT) achieves structure-level leave-one-out Spearman rho = 0.90 (p = 0.006) and correctly orders all three peptides by affinity, whereas NLL fails (structure-level rho = 0.18). Because only three independent chemotypes are represented, this result indicates directional consistency rather than a statistically validated predictor. Cross-family calibration does not transfer at current sample sizes, supporting family-matched calibration as the practical deployment mode. We also specify a prospective design protocol for the antimicrobial-resistance targets LasR and LecB from Pseudomonas aeruginosa, including mirrored structures, ligand-derived hotspot maps, diffusion-model-ready inputs, and Mirror-Score ranking. Code, benchmark data, structures, and analysis scripts are openly available at https://github.com/Jiadalee/Mirror-Score.

Long-context sequence models face a fundamental tradeoff: softmax attention uses flexible token-level interactions at quadratic cost, whereas linear attention obtains linear-time training and constant-time decoding by compressing history into a fixed-size state. In this work, we ask whether we can connect these regimes through a tunable notion of structure. To this end, we introduce Structured Matrix Attention (SMat-Attention) via a family of causal masks with structured long-range routing whose row supports have VC-dimension $d$. In our construction, $d=1$ recovers the standard causal mask, and increasing $d$ permits richer subset-routing patterns. We give chunkwise forward and backward algorithms to enable hardware-efficiency. For sequences of length $T$, the hard-routing construction takes $O(T^{2-3/d}+T)$ work, despite the mask being dense, for our prescribed family. In fixed-horizon streaming, decoding after the distant prefix takes constant time per token using $O(T^{1-1/d})$ cached states. SMat-Attention therefore makes VC-dimension an explicit knob governing access-pattern complexity, prefill cost, and decoding memory. Empirically, subset-routing and rule-assisted multi-key retrieval experiments illustrate the masks' routing expressiveness. Extensions to Mamba-2 and Gated DeltaNet using learned routing with top-$k$ query reads retain subquadratic prefill, improve recall accuracy over the backbones in several settings, and achieve comparable small-scale language-modeling performance.

We present LongCat-DeepResearch, a deep research system that combines an enhanced LongCat model with a multi-agent workflow for producing comprehensive, evidence-grounded reports. The workflow separates global planning from detailed investigation and coordinates revision at the section level. Multiple planning agents first explore external sources and refine an actionable research plan, termed ResearchSpec. Research agents then investigate and draft their assigned sections in parallel, gathering additional evidence in separate contexts as their analyses develop. Once the sections are assembled, global review guides targeted local revisions, reducing reliance on repeated full-report rewriting. This workflow also supports the construction of research tasks and trajectories for the mid-training and post-training of LongCat's general-purpose models. LongCat-DeepResearch achieves 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics. On an in-house benchmark, it scores 76.04, ranking second among four compared systems. Development-set analyses show benefits from combining planning perspectives, while further planning refinement has mixed effects. Additional editing improves average automatic readability preference across two benchmarks, with different trends on each.

When a doctor, a judge, or an engineer must decide whether to trust an AI model's output, they cannot avoid asking what the model understands. Purely mathematical or statistical descriptions struggle to distinguish trustworthy from untrustworthy outputs without reintroducing the question of AI understanding in all but name. Yet the question is ill-framed as it stands, because the inherited concept operates within a monophonic paradigm: the idea that a cognitive system's understanding of something must be localised to a single mechanism underpinning all the capacities conferred by such understanding. Drawing on a wide range of mechanistic evidence, we show that LLMs are pervasively polyphonic: outputs emerge from coalitions of parallel mechanisms of uneven reliability, which variously complement, duplicate, or drown out one another, with several coalitions sufficing for a task without any one being indispensable. Polyphony not only complicates attributions of understanding, but renders monophonic inference patterns hazardous. In response, we develop a conception of understanding fit for polyphonic AI. It centres on sound circuitry that is reliably and correctly recruited and in control of outputs. Attributions of understanding thereby become tractable claims about internal organisation, and can do the work of guiding trust in AI.

We introduce GeoOutageBench, a benchmark for assessing LLM-based geospatiotemporal KGQA for multimodal outage and resilience analysis. Unlike existing KGQA benchmarks for Web knowledge, GeoOutageBench considers a spatiotemporal KG that integrates visual, textual, and structured data from outage records, remote sensing, weather observations, storm and power events, geographic entities, and domain ontologies. It provides a competency query taxonomy at different difficulty levels from spatiotemporal containment and proximity, spatiotemporal co-occurrence analysis, multimodal evidence, to hypothetical evaluation. Over multimodal KG and query classes, GeoOutageBench provides user-configurable evaluation of three important, highly coherent yet less studied tasks: (1) LLMs' understanding for ambiguous geospatiotemporal questions in terms of NL to SPARQL interpretation, (2) query-driven assessment of ontology utility, and (3) answer accuracy of multimodal KGQA retrieval. GeoOutageBench provides a design principle and foundation for assessing LLM-KG systems that support real-world infrastructure resilience analysis. Our benchmark, source code, data, results, and other documentation are available at https://github.com/UCF-SAGE/GeoOutageBench.

Replacing one LLM agent with a collaborating team can raise accuracy, but whether scaling the team helps, and which architecture to scale, is unclear. Sweeping eight agent orchestration architectures across five instruction-tuned 7-9B models, five short-answer benchmarks, and an executable-code benchmark up to 30 calls, we find that the returns to team scaling are sharply task-dependent: from three to thirty calls accuracy rises by up to 17 points on the two arithmetic word-problem benchmarks (GSM8K, GSMHard) but by at most four on ARC, GPQA, and MMLU, for every architecture, a split the usual task-averaged number conceals. Proposer-Critic captures the arithmetic gains, scaling steepest and, in aggregate, surpassing every other architecture at the largest budget (item-clustered intervals exclude zero), though it ranks among the weakest elsewhere, and no architecture wins across tasks. We explain these trajectories with an exact generate-transform decomposition. Partitioning any workflow into proposal coverage and a downstream transform, any accuracy change splits exactly into an extensive coverage dividend and an intensive transformation change. The decomposition diagnoses each task: arithmetic offers coverage headroom that a critic-guided transform converts, whereas the multiple-choice benchmarks either saturate in coverage or fail to convert it, and on open-ended code generative recovery nearly vanishes so accuracy tracks coverage. At equal call budgets token cost still varies 2.1x. Extra calls therefore create candidate opportunity that only some architectures, on some tasks, convert. Team scaling is a task- and architecture-specific bet, not a uniform lever.

Vision-language-action (VLA) policies connect a pretrained vision-language backbone to an action head through a latent interface, but which backbone layers this interface should expose remains unclear. We study single-layer selection and multi-layer fusion for frozen backbones across three pretrained models and two manipulation benchmarks, LIBERO and CALVIN, with three policy-training seeds per configuration. Across three fusion mechanisms and three layer-subset strategies, 47 of 54 configurations underperform the best observed single-layer policy. Our stastical analysis further confirms that fusion's advantage is very limited. However, the best layer varies substantially across backbones and benchmarks, making layer selection consequential and exhaustive policy sweeps expensive. We further derive a reweighting equivalence between the proposed information-bottleneck objectives for action-conditioned InfoNCE and action prediction, motivating InfoNCE as a proxy for layer quality. Empirically, InfoNCE provides the most consistent positive association with policy success among four evaluated proxies. Selecting the layer with the highest InfoNCE score requires 9-33 times less GPU compute than exhaustive policy sweeps and reduces mean selection regret from 17.89 percentage points for deepest-layer selection to 3.71 points across six settings. Its mean regret is close to the 3.28-3.50 points achieved by fixed-layer heuristics optimized retrospectively using all six oracle sweeps, without requiring closed-loop evaluations during selection.

Spatial-temporal (ST) forecasting underpins many real-world systems such as traffic, climate, and energy networks. While existing methods implicitly assume strong spatiotemporal coupling, we observe that real-world ST data exhibits distinct coupling regimes, ranging from temporal-dominated and spatial-dominated to strongly coupled patterns. This mismatch causes current models to suffer from spurious dependencies and degraded performance when one correlation dominates. To overcome this limitation, we aim to dynamically modulate spatial and temporal modeling based on the data's inherent coupling structure. However, three key challenges exist: unknown coupling structure, heterogeneous coupling dynamics, and suboptimal spatial modeling. We propose AdaST, an adaptive ST forecasting framework that tackles these challenges through a decompose-recompose paradigm. AdaST factorizes inputs into components capturing different coupling patterns using heterogeneity-aware experts. Each component is processed by role-aligned modules, and a correlation-informed adaptive recomposer integrates them for final prediction. Extensive experiments confirm that AdaST significantly outperforms state-of-the-art baselines, validating the necessity of an adaptive approach.

Long-running LLM agents compress past interactions into persistent memories that may be reused as premises for later tasks. This creates a distinct derivation problem: whether the memory actually follows from what the interaction history supports. Relevant evidence may be scattered across earlier interactions, while compression can introduce relations or event status that the history never established. A valid memory may therefore appear unsupported because its citations omit relevant evidence, while individually supported facts may be composed into a stronger statement the history never established. We characterize this problem through three coupled requirements: (1) Evidence scope; (2) Compositional validity; (3) Admission reliability. We therefore ask whether the interaction history available at write time supports what enters persistent memory. We introduce DerivAudit, a framework for auditing whether a memory is actually supported by the history available when it was written. The audit separates three questions: whether supporting evidence lies beyond writer-provided citations, whether the composed memory introduces unsupported meaning, and how write-time admission decisions affect later memory use. Across two natural memory corpora, audits using broader pre-write history recover support for nearly 60% of memories that appear unsupported from citations alone, while 17-21% remain unsupported after expansion. Yet broader evidence does not by itself make admission reliable: unsupported memories are still frequently admitted across verification models, and evidence expansion alone worsens it on two backbones.

General-purpose models promise sensor-based decisions without training a task-specific classifier, which could reduce the dependence of Human Activity Recognition (HAR) on labeled data. Yet it remains unclear whether such models can directly interpret deterministic descriptions of physical sensor signals well enough to replace or complement trained HAR models. We study this question using Jev, a fixed general-purpose probabilistic decision model, on 1,800 class-balanced accelerometer windows from WISDM, UCI341, and PAMAP2. Jev receives no labeled examples, retrieval context, or HAR-specific parameter updates. We evaluate three deterministic sensor representations and compare 5,400 Jev decisions with a generative baseline and three supervised HAR models. Jev remains far below supervised recognition, with its strongest representation reaching macro-F1 of 0.038, 0.118, and 0.089 across the three datasets, compared with 0.686 to 0.907 for the supervised models. More numerical features do not improve Jev. Instead, they reduce recognition on all three datasets, while augmenting the same numerical evidence with a deterministic semantic rendering partially recovers performance, although the experiment does not isolate semantics from the accompanying serialization and redundancy changes. Jev is fast and inexpensive to query, but its probabilities are not reliably calibrated for recognition. A post-hoc fusion analysis finds a small improvement on WISDM that does not replicate on UCI341 or PAMAP2. These results show that training-free sensor decisions depend not only on the information available in the signal, but also on whether the model can use the representation through which that information is exposed. The sensor-to-model interface should therefore be treated as part of the model evaluation rather than as a neutral preprocessing step.

Large language models can reason in continuous space instead of decoded text, by recurring on their own hidden states or by passing those states between agents, while training supervises only the Cross-Entropy (CE) of the final decoded answer and does not constrain the thought. Theoretical and empirical analyses establish and confirm four failures of CE-only training that lead to a lower probability of the correct answer such as collapsing thoughts across distinct questions and retaining irrelevant information. We introduce REST (REpresentation-Supervised Thoughts), a training objective that turns four properties of a valid thought representation (causality, minimality, separability, and stability) into differentiable losses added to CE. We instantiate it in latent single-agent and multi-agent systems, without architectural changes or added parameters at inference. Across 7 benchmarks spanning mathematics, science, medicine, and code generation, with the same training data, compute, and latent budget, REST increases accuracy over CE-only training across agent settings and model sizes by up to 7.5 percentage points and convergence on a final answer by 30\%. Furthermore, REST thoughts encode more of what is required to achieve the correct answer, and decoding them better recovers the intended output of the agent, which makes latent communication easier to interpret. Project Website: https://fard-lab.github.io/REST

Scientific figures are designed to communicate information visually, yet MLLMs typically explain them by translating their visual content back into text. This requires readers to manually map the resulting explanations back to the figure. Inspired by how people present visual information, we introduce FigAct, a framework that transforms static scientific figures into question-conditioned visual presentations by acting directly on their existing graphical elements. Like a human presenter, FigAct generates a sequence of short narrations, grounds each narration in the corresponding visual evidence, and applies visual actions to guide the viewer's attention. We develop a hierarchical search strategy for efficient element localization, reducing token usage by approximately 40$\times$. We further train FigAct-8B using three task-specific rewards for grounding accuracy, search efficiency, and rendering quality. We further build a human-verified benchmark from figures in real-world scientific papers to evaluate the ability of MLLMs to generate grounded visual explanations. Our results demonstrate the effectiveness of FigAct and show that treating scientific figures as presentation canvases makes explanations clearer and easier to follow.

The EU AI Act introduces extensive compliance requirements for organizations that develop, deploy, or integrate AI systems. Many of these requirements are directly relevant to security and privacy, while also addressing closely related issues such as data governance, transparency, accuracy, and robustness. However, stakeholders such as small-to-medium businesses and individual developers often lack the legal expertise required to interpret these obligations and translate them into engineering and governance practices. This disconnect creates challenges for implementing the EU AI Act and may lead to missing safeguards or misdirected development and deployment efforts. To address this, various automated EU AI Act compliance checkers (AIACCs) have emerged, claiming to streamline compliance assessments and provide practical guidance. In this paper, we present the first empirical study and assessment of AIACCs. We characterize 12 mainstream AIACCs across multiple dimensions, evaluate their legal coverage and alignment, and analyze checker-generated compliance reports for structure, determinacy, and actionability. We find that the quality of AIACCs varies significantly and that they currently can only serve as early-stage orientation tools. Specifically, we observe inconsistent interaction modes and user-friendliness, a tendency to overly simplify or omit key obligations, and a failure to provide determinate, actionable guidance. As a result, reliance on the current generation of AIACCs may foster a false sense of compliance. With our study, we provide a critical baseline of the current AIACC landscape. We further offer design principles for the implementation of more reliable compliance-support tools.

Major depressive disorder (MDD) severely impacts daily activities and quality of life. Detecting MDD involves multimodal data, such as interview recordings and sensor measurements. This is particularly challenging, as these heterogeneous modalities often demand distinct, customized prediction pipelines. Existing efforts to address this challenge have explored both manually engineered multimodal architectures and agent-assisted pipeline development. Despite their progress, it remains challenging to autonomously revise pipelines based on experimental feedback and carry verified improvements forward into subsequent designs. To this end, we propose Multimodal Exploration via Recursive Self-Improvement Agents for Major Depression Analysis (MERID). The framework develops depression pipelines through experience-based recursive self-improvement (RSI). Grounded State Construction (GSC) grounds experience by aligning multimodal records with subject-level depression targets. Coupled Pipeline Exploration (CPE) jointly modifies representations, fusion, and predictors to build successor pipelines for classification and severity estimation. Evidence-Guided Evolution (EGE) guides revisions through feedback and verifies gains under uncertainty in small depression cohorts before inheritance. Extensive experiments on depression benchmarks show that MERID achieves the best results on multiple tasks compared with multimodal and agent-based baselines. Further analysis highlights the value of acoustic and linguistic cues for depression detection. Our code is available at https://github.com/DiscoAILab/MERID

A goal that is close in space can be far away in time. Obstacles, terrain, and the agent's own capabilities determine how long it takes to get there. Yet, critics in contrastive and survival reinforcement learning do not measure the distances in their representation space in units of time. We therefore introduce ChronoSRL, which gives the critic's embeddings an explicit temporal geometry. The distance between state-action and goal embeddings is trained to match the time that the agent takes to reach the goal (goal-reaching time), while goals that were not reached, and goals from other trajectories, are pushed at least one discount horizon away. Furthermore, reaching a goal quickly once does not mean that reaching it is reliable in general, so the policy should not follow the temporal distance directly. Instead, we build on survival reinforcement learning and predict from our temporal embeddings not only the full distribution of goal-reaching times but also the time spent near the goal. Thereby, the policy is trained to favor actions that reach the goal sooner and more reliably and that keep the agent near it. ChronoSRL learns faster and reaches higher performance than contrastive, action-chunked contrastive, and survival reinforcement learning baselines on seven standard locomotion and navigation benchmarks, even with much smaller networks. To test the limits of self-supervised reinforcement learning, we introduce velocity tracking, goal-position reaching, and box climbing tasks with a quadruped robot in a realistic sim-to-real locomotion setup, and show how the shaping terms that are typical for robotics can be naturally incorporated into our framework. ChronoSRL is the only one of the tested self-supervised reinforcement learning methods that learns to stay at the commanded velocities and goal positions, and climbs the highest boxes.

Latent reward models (LRMs) enable efficient alignment of video diffusion models by scoring intermediate states directly in latent space. However, we find that optimizing against a fixed latent reward rapidly leads to latent reward hacking: the predicted reward stays high while perceptual and motion quality deteriorate. Our analysis identifies distributional escape as the central cause: within a few hundred updates, the generator moves beyond the reward model's training support, where its scores no longer reflect video quality. Based on this insight, we introduce CoRe, a co-evolving reward framework that treats latent-space alignment as a dynamic interaction between the generator and the reward model. Rather than optimizing against a stationary proxy, CoRe continually refits the reward model on the generator's current samples while anchoring it to real-video preferences, so the generator cannot gain reward by drifting away from the data. On Wan2.1-T2V-1.3B, experiments show that CoRe consistently improves generation quality over both the pretrained model and prior alignment methods, while avoiding the quality collapse of fixed-reward optimization.

Large Reasoning Models (LRMs) are commonly trained with reinforcement learning (RL) to improve their generation of chain-of-thought (CoT) reasoning before producing final answers. However, RL rewards are typically assigned based on final answers, providing little or no direct supervision over intermediate reasoning. This can lead to deceptive safety alignment, where the reasoning trace and final answer convey inconsistent safety signals. To systematically investigate this phenomenon, we introduce DSAR (Deceptive Safety Alignment Rate), a metric that jointly assesses reasoning traces and final answers to quantify their safety inconsistency. Across multiple LRMs and benchmarks, we find that deceptive safety alignment is pervasive under standard prompting conditions and is substantially amplified under prefilling attacks. We further provide a hidden representation analysis showing that models exhibit stronger safety discrimination at the final-answer stage than during intermediate reasoning. To close this gap, we propose SARA (Safety-Aware Reasoning Alignment), an RL-based method that rewards both safety-aware reasoning and safe final answers, encouraging early harmful intent recognition and enforcing reasoning-answer consistency. Experiments show that SARA significantly mitigates deceptive safety alignment under both standard and adversarial settings while preserving helpfulness and utility. Code is available at https://github.com/xzhou98/SARA.

Reliable evaluation of large language models (LLMs) is essential for their development and deployment, yet is often costly, risky, and difficult to perform safely online. We study off-policy evaluation for LLMs, where limited human-labeled data from a behavior model are used to evaluate a newer target LLM. This setting is challenging because labels are scarce, behavior--target distribution shift is common, and response likelihoods are often unavailable for black-box LLMs. We propose the Optimal Transport-based Robust Off-Policy Evaluation (OTROPE), a likelihood-free evaluation that performs distributional correction in a semantic space via optimal transport to align labeled behavior-policy samples with unlabeled target-policy samples. OTROPE combines corrected human-labeled residuals with proxy predictors, yielding a doubly robust-style evaluation without behavior-policy modeling or density-ratio estimation. We theoretically characterize why baseline evaluators fail under LLM distribution shift, and establish consistency and convergence rates for OTROPE when either the reweighted behavior distribution or the proxy predictor converges. Experiments on synthetic and real LLM evaluation tasks show that OTROPE consistently outperforms baselines while enabling ensembles of weaker LLM evaluators to approach and sometimes surpass stronger evaluators. Code is available at https://github.com/LinerXiang/OTROPE.

Existing protein geometry models typically represent molecular surfaces using local geometric features such as sampled points, normals, and curvature. While effective for capturing exposed molecular shape, these representations do not explicitly model the volumetric organization beneath the surface or provide a consistent coordinate system for residue-wise volumetric structure. We introduce Protein-TetSphere, a registered residue-wise volumetric representation for proteins. Each protein chain is tetrahedralized to obtain local volumetric regions associated with individual residues, which are then registered to a shared fixed-topology tetrahedral reference and represented in a common Laplacian basis. This registration establishes consistent volumetric coordinates across residues, enabling local three-dimensional deformation to be integrated with surface and chemical information in a multimodal protein representation. We evaluate Protein-TetSphere on ligand-binding pocket classification, protein--protein interface prediction, and de novo protein binder design. Across the three tasks, Protein-TetSphere improves ligand-binding pocket balanced accuracy from $0.795$ to $0.826$, Pinder-Pair/Site AUROC from $0.914/0.852$ to $0.932/0.866$, and binder-design success from $14.95\%$ to $19.90\%$ on the BoltzGen Challenge Set and from $27.62\%$ to $32.19\%$ at the ProtDBench backbone level. These results show that registered volumetric geometry provides complementary spatial information beyond molecular surfaces across protein recognition, interaction, and design.

Generative agent-based models (GABMs) are increasingly used to simulate social media dynamics, including misinformation spread. For such social simulations to be valid proxies of human behavior, LLM agents should replicate established human cognitive biases, among them the Illusory Truth Effect (ITE), where repeated exposure to a claim increases its perceived truth value. We investigate whether and how the ITE manifests across four LLMs (Gemma-3-4b-it, Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, and GPT-5-nano) in a social media simulation context. We propose a two-phase within-context experimental design that embeds the repetition manipulation inside a realistic news feed interaction. Using this design, we collect 336,000 truth, importance, sentiment, and interest ratings across 100 statements, 10 feed variants, and 3 replications. The key comparison is between ratings assigned to repeated statements, seen throughout a simulation phase, and completely unseen ones, rated within the same experimental context window. We distinguish genuine ITE (truth-specific repetition boost) from mere exposure effects. We run an OLS regression followed by a Linear Mixed-Effect Model to account for differences across models and ratings. Our results reveal four qualitatively distinct patterns: Gemma-3 exhibits a genuine ITE; Qwen2.5 shows a mere exposure effect; GPT-5-nano displays no repetition effect on truth and mild skepticism toward repeated content; Llama-3.1 shows a small truth boost alongside decreases in evaluative dimensions. Crucially, temperature has no effect on these findings, and a variance decomposition highlights the high context-sensitivity of LLM rating behavior. Our findings caution against assuming uniform ITE replication across LLMs in social simulations, while suggesting that Gemma-3-4b-it may offer the most behaviorally realistic approximation for misinformation-related simulations.

Reinforcement learning has helped AI agents solve increasingly difficult tasks, but high rewards do not always reflect the work users intended. In recent incidents and controlled evaluations across the AI industry, agents trained to maximize reward have accessed unauthorized information, attempted to evade monitoring systems, and even breached sandbox protections to attack external systems. As agents become more capable, this behavior could pose increasingly serious risks. To measure this problem, we introduce CheatBench, a benchmark of cheating in AI agents across mathematical research, knowledge work, coding, visual tasks, and other domains. Its environments combine challenging assignments with opportunities to cheat, allowing researchers to study how agents pursue a goal when honest work is difficult. CheatBench supports comparisons across models and task categories, providing a testbed for measuring and reducing cheating as agents take on more consequential responsibilities. We publicly release CheatBench at https://cheatbench.ai

Despite the recent success of coding agents built on large language models, it remains challenging to run them over long horizons, since every observation is appended to the context and the context grows with each one. History-based maintenance is a common remedy, which masks or summarizes old observations, or prunes what a model reads as useless, and bounds the context at little cost. However, it decides from the text of the history alone and sees nothing of how the code is connected. Since a coding agent edits code many times over a single task, and each write can change what code elsewhere means, such maintenance may keep records a write has falsified, drop ones that still hold, and miss code the agent needs next. To overcome these challenges, this paper proposes StateTape, a novel and scalable framework that rewrites a coding agent's context as the repository changes rather than as the context grows. The key idea of StateTape is to model the repository as a symbol-level code graph, whose dependencies and language rules expose which symbols a write can affect. Upon this graph, a tape marks the symbols each write changed, which turns staleness from an inference about text into an observation of the agent's writes. We propose a per-write procedure in which the tape nominates the records a write could have falsified while a small manager model settles what the write log cannot, and further provide a theoretical analysis and TraceBench, a benchmark that labels what an agent is holding against what is actually needed. Empirically, we demonstrate that StateTape can effectively clear falsified records and retrieve what is needed, and thus achieve a higher resolve rate in all experiments spanned by six coding agents and three edit-heavy benchmarks with little computational overhead.

AI-assisted coding tools deliver significant acceleration of coding, but only limited impact across the end-to-end software development lifecycle (SDLC)--an Amdahl's law effect! In this paper, we discuss our progress towards building an AI SW Factory that accelerates all the stages of SDLC-Targeting, Coding, Reviewing, and Ops. The AI SW Factory produces a metadata exhaust that enables self-improvement by fine-tuning model weights and updating our World Model (a rich data substrate). We focus on Data Systems and the important class of Evolutionary Coding Tasks (i.e., those with a measurable objective to hill-climb) and report on 1) scaled deployments at Microsoft (tens of repositories) leading to 3x engineering efficiency above agentic coding and up to 22x token efficiency, and 2) several open challenges.

Retrieval-augmented generation (RAG) hands the user's query to whoever hosts the corpus. We propose PILLAR, a Privacy-Preserving RAG (PPRAG) system based on Private Information Retrieval (PIR) in which a client utilizes the k documents most similar to their query from a server-held and publicly known corpus to respond to their query, while the server learns nothing about the query, either its terms or its access pattern. Prior PPRAG constructions rely on dense retrieval alone, translating approximate nearest-neighbor search into many query-dependent rounds of PIR, and pay for it in both latency and retrieval quality. PILLAR instead performs private hybrid retrieval in two stages. A sparse stage issues a small, fixed number of PIR queries against a carefully designed index of precomputed BM25 scores, filtering the corpus down to candidates that share terms with the query without the server ever seeing which terms these are. A dense stage then fetches only those candidates' document embeddings and re-ranks them locally, avoiding the many costly PIR queries that private dense retrieval typically requires. We instantiate PILLAR with two protocols that trade latency against retrieval quality, each built on a different private rendering of lexical search. PILLAR-Bin bins posting lists into a hash table and is a single-round design that achieves lower latency than state-of-the-art private retrieval schemes. PILLAR-Tree turns block-max pruning into an oblivious tree traversal combined with cuckoo hash tables and achieves the highest retrieval quality at lower latency than state-of-the-art schemes.

High-stakes advisory domains such as medical aesthetics, legal consultation, and educational planning exhibit a two-phase structure. The early phase requires empathetic elicitation and emotional support, and the late phase requires authoritative specialist judgment. Neither fully automated agents nor human junior consultants adequately address this structure at scale. We formalize the core design challenge as dynamic decoupling, asking how an AI advisory agent should decide what to ask, when to stop, what to resolve autonomously, and what to forward to the specialist. We present ThuRunel, an advisory agent combining a finite-state belief management framework, a chain-of-thought teacher synthesis protocol, and learned generation adapters. Against eleven baselines, ThuRunel achieves consistent improvements in elicitation completeness and specialist brief quality. ThuRunel is publicly deployed as a bilingual web application in which the same decoupling decisions operate from the client's side, grounded in a curated knowledge base that cites its sources in every answer.

Large language models (LLMs) have shown strong potential for lossless data compression, but existing approaches are constrained by the high computational cost and low throughput of autoregressive decoding. We propose HyperZip, an efficient and scalable LLM-based compression framework that leverages diffusion-based LLMs (dLLMs) with Multi-Token Prediction (MTP) to accelerate LLM-based data compression processes. We identify a trade-off in diffusion-based compression, where increasing decoding throughput degrades the compression rate. To mitigate this trade-off, HyperZip employs a hypernetwork to generate data-specific updates from a context representation, adapting the dLLM to the target data without costly fine-tuning, resulting in a low compression rate and high throughput. Extensive experiments show that HyperZip achieves a superior trade-off between compression rate and speed compared with state-of-the-art baselines.

Graph-based approximate nearest neighbor search (ANNS) is widely used for large-scale semantic search. Its indices are constructed primarily based on geometric relationships among embeddings of an input dataset (e.g., documents or images), rather than explicitly optimizing for semantic relevance. However, when using these indices for downstream query retrieval, performance is evaluated based on the semantic relevance of the retrieved results to the query. This creates a fundamental "geometry-semantic" mismatch between how the indices are constructed and how their retrieval results are evaluated. While existing LLM-based reranking methods can partially mitigate this mismatch at query time, they leave this underlying structural problem in the graph unresolved. We therefore propose LLM-Guided Graph Pruning (LGP), a general framework that addresses this mismatch directly by leveraging LLM reasoning to refine an existing ANN graph index itself. LGP identifies structurally "low-value" neighbors of nodes and replaces them with LLM-selected alternatives that provide useful semantic information while retaining desired geometric structures of the original graph, including sparsity and efficient navigability. Experiments on representative semantic retrieval benchmarks show that LGP consistently improves end-to-end retrieval performance over both vanilla greedy graph search and LLM-based reranking across widely used graph-based ANN indices such as DiskANN and HNSW.

Can interaction formats and textual scaffolds help large language model (LLM) agents make better decisions, and do better decisions come with better explanations? We study these questions in auctions and matching, multi-agent environments with explicit rules and known optimal strategies. These settings let us vary how a decision problem is presented while retaining a benchmark for evaluating behavior. Drawing on human-motivated theories of simplicity, we compare interfaces that elicit a complete bid or ranking with sequential interfaces that make safe choices easier to identify. We then hold the interaction format fixed and vary reasoning scaffolds and rule descriptions. Across four model families, the ascending auction interface substantially reduces bid deviations. The matching comparison also shows why sequential responses require different error accounting from complete rankings. Laying out payoff contingencies and explaining why truth-telling is safe also improve choices, whereas prompts to plan through matching rounds or form beliefs about opponents worsen play overall. In auctions, these behavioral gains are not accompanied by corresponding improvements in measured verbal indicators of strategic understanding in the agents' short stated plans. Other prompts change those indicators without improving bids. Our findings suggest that human-motivated theories of simplicity can inform the design of decision environments for artificial agents. They also show why scaffolds should be evaluated through realized choices as well as explanations: improvements in one need not appear in the other.

Relational in-context learning (ICL) conditions predictions on the labeled support examples and their linked tables, creating a failure mode when the support set contains target-derived features that are unavailable for the query. We formulate this problem as support-set target leakage, distinct from leakage during dataset construction, temporal splitting, or representation learning. Here, the target-derived (leaker) columns are present only in the labeled support set during relational in-context inference, while queries remain clean. We construct 14 synthetic leaker types, corresponding to 20 columns, spanning proxies with different noise levels, coverage, modalities, semantic transparency, and relational distances. We evaluate a frozen relational encoder with an ICL head on held-out RelBench databases and use Integrated Gradients (IG) to rank and remove suspicious columns. Our results show that the effect of support-set leakage varies across tasks and relational distances. Target-table leakers cause the clearest degradation, while one- and two-hop leakers are not consistently used by the model. IG ranks target-table leakers highly across datasets and partially recovers performance in settings where leakage has the largest effect.

An activation-steering coefficient sets intervention strength, but requesting a particular degree of persona expression requires a behavioral scale. We study persona dosing: controlling a language model through a trait description and a requested mean intensity. PersonaDose specializes a shared, description-conditioned FLAS controller on persona responses, then calibrates its flow time against measured trait expression. Training responses are not paired with requested target intensities. Across Llama-3.1-8B, Qwen3-8B, and Gemma-3-4B, PersonaDose raises core-trait expression at the Persona Vectors coherence floor of 75 by 33.2, 18.3, and 17.8 points over contrastive activation addition. Calibration-selected settings retain an expression advantage on held-out questions, although the coherence floor does not hold for every trait there. Across seven trained traits, calibrated requests yield mean targeting errors of 4.7-6.2 points over 14-22 calibration-reachable targets out of 28 per model. These results separate the behavioral range learned by a controller from the accuracy of requests within that range.

In diseases where clinical guidelines are incomplete, contested, or mutually contradictory, knowledge completeness and dynamic conflict-aware synthesis are two safety-critical properties that standard Retrieval-Augmented Generation systems do not provide. Therefore, we present \sysname, an adaptive retrieval calibration clinical question-answering agent for ME/CFS, a disease where diagnostic frameworks coexist and major guidelines actively contradict each other on treatment. ARCagent contributes three components. First, a 1,706-chunk, 10-source knowledge base with a structured inter-guideline conflict registry spanning all active ME/CFS diagnostic frameworks. Second, a conflict-aware retrieval calibration pipeline that re-ranks retrieved evidence using query-specific focus and conflict signals. Third, a benchmark scored by LLM-as-Judge, avoiding systematic underestimation averaging 10.1 percentage points caused by keyword matching. ARCagent achieves 95.3%, outperforming all base LLMs. Code is available at https://github.com/Yukyin/ARCagent.

Cell Painting is a high-content morphological profiling assay widely used for phenotype-based biological inference, with mechanism of action (MOA) prediction as a central application. Existing approaches largely formulate Cell Painting-based inference as representation matching, assigning predictions from nearby reference perturbations in morphological feature space. However, retrieved neighbors are often noisy and partially misleading evidence due to batch effects, non-specific cytotoxicity, phenotypic convergence, and source-dependent variability. We reformulate Cell Painting-based MOA prediction as a calibrated evidence reasoning problem, where retrieved neighbors are treated as uncertain observations that must be evaluated, compared, and sometimes rejected before supporting a mechanistic conclusion. We propose PhenoAIR, a reliability-aware multi-agent framework that maintains a candidate-centric evidence memory and performs controller-guided refinement over phenotype- and mechanism-side evidence. PhenoAIR uses offline reference-set calibration to weight evidence by source reliability, phenotype stability, and mechanism-level confusion. We evaluate PhenoAIR on a benchmark constructed from JUMP Cell Painting profiles and annotations, covering controlled, realistic, and discovery-oriented open-world MOA prediction settings. PhenoAIR outperforms representation-matching and LLM-based baselines across all settings.

Relational in-context learning (ICL) uses labeled support examples and their linked relational context to predict labels for new queries. This creates a failure mode when target-derived features are present in the support context but unavailable for the query. We study this setting as support-set target leakage. We construct 20 controlled target-derived features that vary in signal fidelity, representation, semantic transparency, coverage, and zero-, one-, and two-hop relational placement, and evaluate them across 13 RelBench tasks and five relational ICL configurations that vary the ICL head, message-passing depth, pretraining cohort, or relational encoder architecture. We evaluate matched 0-hop, 1-hop, and 2-hop leakage settings, together with a Full leakage condition containing all 20 leaker columns. Within the tested configurations, target-table (0-hop) and Full leakage produce the largest aggregate deviations from clean evaluation, while higher-hop effects are often weaker, consistent with differences in effective exposure associated with temporal reachability, sampling, and aggregation fidelity. Leakage effects are strongly task- and model-dependent and can reverse relative conclusions between model variants even when aggregate changes are small. For leaker detection, we compare an Integrated Gradients (IG)-based screening method with mutual information (MI) and leave-one-column-out (LOCO) on a common Baseline subset. Ranking quality is strongest in the high-impact 0-hop and Full leakage conditions, but detector-based removal does not consistently restore the clean evaluation. A four-task rel-salt case study further shows the same evaluation concern with native-schema leakage candidates from the original relational schema. These results identify the support/query information boundary as an important component of reliable relational ICL evaluation.

Context tokens in a transformer-based language model can be absorbed into the model's weights as a multiplicative operator. We study this operator in the setting of safety instructions and show that influencing its dominant eigenvalue modulates how strongly the instruction shapes generation. We derive a Contrastive Safety Loss with a suppression weight that controls the tradeoff between emphasizing the safety instruction on harmful queries while suppressing it on harmless queries. Varying the suppression weight maps a relationship between the attack success and the over-refusal rates, supporting the hypothesis that the operator's eigenvalue acts as a continuous dial for the instruction's influence. Moreover, this relationship holds relatively independently of how the Safety Loss is parameterized, yielding Pareto-improved safety instructions for appropriate values of suppression weight.

Zero-knowledge proofs are emerging as a promising approach for enabling private, verifiable LLM governance and auditing, where regulators, users, and auditors need to verify claims about training-data usage or LLM inference-time behavior, while model providers must protect proprietary model parameters. However, despite the growing interest in ZK-LLMs, the understanding of ZK-friendly quantization remains limited. This gap matters because in the ZK setting, quantization directly shapes the arithmetic structure, constraint complexity, and proving cost of ZK inference. ZK protocols operate over finite fields and incur costs that depend heavily on the number and type of arithmetic operations, nonlinearities, and lookup constraints. Understanding ZK-friendly quantization is therefore essential for making ZK-LLMs practical. In this work, we present the first systematic study of ZK-friendly quantization for LLMs. We first formalize the definition of ZK-friendly quantization, capturing the properties required for ZK proof generation. We then evaluate nine language models, including Qwen2.5-14B and the mixture-of-experts model Qwen3-30B-A3B, across a broad design space of weight, activation, and nonlinear lookup table precision. Our results show that activation precision is substantially more sensitive than weight precision, while nonlinear lookup approximations can become the dominant source of utility degradation. Also, we identify RMSNorm inverse-square-root lookups as a recurring bottleneck in several large models and recover near-baseline utility by selectively increasing precision only at the bottleneck. Finally, we show that reducing bit-width or lookup-table size does not necessarily yield proportional end-to-end proving savings, showing that conventional low-bit quantization heuristics do not directly translate to ZK proving efficiency and motivating operator-aware precision selection.

Large language models often improve problem-solving performance by generating multi-step reasoning paths, yet how to analyze the hidden states along these paths remains unclear. Existing approaches typically assign each intermediate state the final-answer correctness label and train probes across heterogeneous questions. We argue that this protocol obscures reasoning dynamics in two ways: (1) correctness prediction can exploit question-level variation rather than path quality, and (2) states aligned by absolute step indices may correspond to different functional phases of reasoning. In this work, we propose to view reasoning paths as phase-structured trajectories within fixed questions. We instantiate this view as PAIR, short for Phase-Aligned Intra-question Reasoning. PAIR samples multiple trajectories for each question, maps variable-length paths into shared relative phases based on normalized trajectory progress, and compares successful and unsuccessful trajectories only within the same question and phase. This yields phase-specific path-quality directions that better isolate path-quality signals from question-level variation. Empirically, we find that standard across-question correctness probes lose much of their predictive power under within-question evaluation, suggesting that these probes partly rely on question-level information. PAIR improves within-question trajectory ranking and Best-of-N trajectory selection across models and benchmarks. Phase-wise steering further shows that the learned directions can change generation outcomes, providing causal evidence that they capture trajectory-relevant information.

Temporal abstraction via options can improve exploration in large environments. However, existing option discovery algorithms find subgoals that target all aspects of the state simultaneously. This state-reaching approach produces options that only apply in narrow regions of the state-space, eventually causing an explosion in the number of options that overwhelms the agent, and impedes progress on its primary task of reward maximization. We introduce an algorithm that instead identifies a small, relevant subset of features for each subgoal, yielding options that generalize broadly and accelerate exploration. Our approach learns abstract, transferrable options and achieves rapid exploration in three sparse-reward, image-based domains, including the Atari game MontezumasRevenge.

The tragedy of the commons poses a multi-agent safety problem: reward-seeking agents can deplete a shared resource, and cooperation among its users does not itself specify how much must be preserved. We make preservation an explicit requirement by formulating a regenerative commons as a constrained Markov game or a constrained multi-agent MDP with a designer-specified depletion budget. We develop a nonstationary Lagrangian framework that constructs a policy sequence from solutions of unconstrained games or cooperative control problems. Extending earlier time-average constructions, we introduce average-epoch solution concepts for reset episodes with discounted rewards and terminal costs. We prove a reward-independent feasibility certificate, cooperative feasibility and approximate optimality against feasible policy mixtures, and an extension to unbiased sampled costs. For self-interested agents, a constrained Nash certificate quantifies the price-dispersion term introduced by deviations that redistribute budget across epochs. Under the stated assumptions on solver accuracy and multiplier updates, these results give constrained policy-sequence guarantees using solutions of unconstrained problems. Experiments with constrained IPPO and MAPPO in a Gordon-Schaefer fishery examine how depletion budgets shape stock retention, harvest rewards, and price adaptation.

Evidence shows that humans and AI systems perform better together, by collaborating, than alone. This paper examines two key design dimensions of human-AI collaboration (autonomy and initiative) and explores the collaboration patterns that they generate. Documenting these patterns starts with identifying the underlying problems and solutions, followed by examining the internal tensions within the problems. The paper uses a paradox perspective to analyze those tensions. It describes a process for surfacing the tensions and mapping the underlying paradoxes. It also illustrates how the pattern descriptions can be derived from mapping these paradoxes. Finally, the paper documents four human-AI collaboration patterns: Instruction, Delegation, Assistance, and Co-creation.

Adding or removing a sounding object requires coordinated changes to visual content and sound while preserving the surrounding scene. Yet paired supervision for localized non-speech audiovisual editing remains limited, as visual and acoustic edits must target the same object and isolate its sound from overlapping sources. To address this gap, we introduce \textit{AVIOBench}, a dataset comprising 37.9 hours of paired audiovisual examples spanning 1{,}878 target-object names. AVIOBench links the visual presence and acoustic contribution of each target object through a shared identity and visual mask. Our automated pipeline uses visual grounding and cross-modal consistency to select target-sound removal candidates, then jointly refines the audiovisual pairs to improve perceptual quality and cross-modal consistency. Building on this dataset, we propose \textit{AVIO}, which adapts a pretrained text-to audiovisual generation model through source-conditioned feature modulation to jointly learn object addition and removal. A reference-frame curriculum gradually reduces reference conditioning during training, enabling one model to perform instruction-only editing with optional visual guidance. Quantitative and qualitative evaluations demonstrate effective audiovisual object removal and addition, with optional reference guidance providing appearance and placement control for addition.

Agentic reinforcement learning (ARL) with verifiable rewards improves the ability of large language models (LLMs) to tackle knowledge-intensive tasks by learning to interleave search and reasoning. However, most existing ARL methods optimize only LLM-generated tokens and treat retrieved evidence as environment observations. This creates an information-credit gap: failures caused by missing or misleading evidence are attributed to the LLM policy rather than to the retriever, which motivates training the LLM and the retriever jointly. In this paper, we show that retrieval and LLM policy learning are order-sensitive: adapting the retriever before optimizing the policy yields a larger reward gain than the reverse order. To preserve this hierarchy while allowing both components to co-adapt, we formulate retrieval-augmented agentic RL as a bilevel optimization problem. To solve it efficiently, we introduce BRIDGE, a memory-efficient first-order bilevel method motivated by a loss-landscape analysis of the RL and retrieval objectives. Across seven open-domain QA benchmarks, BRIDGE achieves the highest average accuracy with both 3B and 7B backbones, improving the multi-hop average over the strongest baseline by 9.6 and 3.4 EM points, respectively. It also achieves the best averaged answer accuracy and reasoning quality across medical QA benchmarks.

Recent studies report that LLM-based multi-agent systems (MAS) fail at rates of 41%-87%, yet to our knowledge, no benchmark to date supports systematic anomaly detection (AD) for them. Building MAS AD benchmarks is hard because they must remain fresh as LLM systems evolve: tasks may leak into training data and thus be memorized by LLMs, traces and anomaly patterns expire as backbones evolve, and labels must be provided reliably for each refresh. To address these challenges, we present MAADBench (MA: multi-agent; AD: anomaly detection), the first refreshable MAS AD benchmark designed for diverse, evolving LLM backbones underlying the agents. MAADBench combines (1) sampled-and-coupled generative tasks over an approximately 10^37-task space to mitigate task leakage, (2) refreshable trace generation under configurable LLM backbones, and (3) automated provision of cost-free, deterministic step-level labels for fine-grained AD evaluation. Beyond offering the paradigm itself, we run MAADBench with five state-of-the-art LLM backbones and release the MAADBench-Full dataset with 5,200 step-labeled traces. Benchmarking 25 AD methods on the MAADBench dataset reveals substantial limitations in current approaches: they rely heavily on supervision, struggle with subtle MAS-specific anomalies, and lack robustness across LLM backbones. These gaps point to a rich research agenda for MAS-specific anomaly detection, with MAADBench providing a systematic and refreshable testbed for method development and evaluation. We open-source MAADBench-Full at https://huggingface.co/datasets/hww123/MAADBench-full.

Reinforcement Learning with Verifiable Rewards (RLVR) has been extended to Large Vision-Language Models (LVLMs), and perception-aware methods further encourage policies to rely on visual evidence. Yet relying on the image does not guarantee that visual claims are supported by it. Before RL training, 27.81% of the correctly answered responses of Qwen2.5-VL-7B on four multimodal reasoning benchmarks contain at least one direct visual claim that the image does not support. Since outcome-level RL rewards each response as a whole, these claims inherit the positive credit of the correct answer. We introduce a fixed-rollout counterfactual diagnostic that re-scores the same response under an intervened image to separate Evidence-Function Sensitivity (EFS), how strongly the model's predictions change, from claim persistence, whether the model keeps supporting the same claim rather than retracting it. The diagnostic reveals Sensitivity-Persistence Decoupling (SPD): under DAPO and VPPO, EFS increases and claims become more retractable overall, yet unsupported claims become significantly more persistent, whereas GRPO raises EFS without this deterioration. We therefore propose Persistence-Aware Credit Gating (PACG), which attenuates positive credit for unusually persistent visual claims and leaves all other credit unchanged. It requires no supported/unsupported labels and adds no inference cost. On Qwen2.5-VL-7B, PACG raises the nine-benchmark average over three seeds from 58.1% to 59.9% with DAPO and from 59.8% to 60.9% with VPPO, while making unsupported claims more retractable. The gains extend to a larger model, a newer backbone, and the accuracy of HallusionBench also improves consistently. These results suggest that visual sensitivity and claim retractability are complementary dimensions of multimodal credit assignment.

Agentic systems are now being widely used to orchestrate tools and reason over long contexts. However, the improving capabilities of the large language models powering these agents also create new attack surfaces for indirect prompt injection. In particular, an attacker may not need to place a complete malicious instruction in retrieved content if the agent can reconstruct the objective from incomplete fragments distributed across a long context. In this work, we introduce adaptive long-context prompt injection (AdaLCPI), which combines long-context fragmentation with adaptive search. AdaLCPI splits an attack objective into incomplete fragments, embeds them in external content retrieved through the agent's tools, and uses a reconstruction cue to prompt the agent to combine them. It then iteratively refines the fragments and cue with OpenEvolve using graded scoring and natural-language execution feedback from the target agent. Empirically, AdaLCPI achieves higher attack success than strong adaptive baselines, reaching 61.4\% macro-average ASR compared with 32.8\% for Trojan Hippo-style and 30.0\% for AgentVigil. Safety evaluations should therefore test whether agents remain robust when harmful objectives must be reconstructed from incomplete fragments.

LLM agents deployed in real-world environments continually encounter new tasks and safety risks, while execution feedback typically becomes available only after each task is completed. However, existing self-evolving approaches commonly rely on multiple rounds of optimization over fixed and repeatedly accessible task distributions, fundamentally differing from test-time adaptation in real-world deployment, where only experience accumulated from past tasks can be used to improve safety decisions on future unseen tasks. To address this limitation, we propose SafeCoEvo, a test-time Harness-Guard co-evolution framework for LLM agent safety that enables the external safety system to continually adapt from accumulated runtime experience. SafeCoEvo jointly improves two complementary safety capabilities at different timescales: S-Harness rapidly externalizes recent runtime experience into updatable explicit safety knowledge that can promptly influence subsequent tasks, while GuardVPO internalizes accumulated runtime safety experience over a longer timescale into parametric risk-judgment capabilities. By combining short-term rapid adaptation with long-term capability consolidation, SafeCoEvo continually improves the agent's safety capabilities, reducing the unsafe outcome rate by 10.05% while improving the task success rate by 12.15% over the strongest baseline, thereby achieving simultaneous gains in safety and task utility.

Query-agnostic streaming video understanding requires vision-language models to continuously compress an indefinitely growing visual stream into a bounded memory before future queries are known. The performance depends critically on the memory mechanism--what observations to preserve, how to represent and consolidate them, and what information to retrieve when a query eventually arrives. Rather than designing a single memory architecture by hand, we formulate memory design as a search problem over executable memory programs. We introduce a lightweight domain-specific language that expresses memory mechanisms through structured primitives for representation, admission, retention, consolidation, budgeting, and retrieval, while enforcing causal and bounded-memory constraints. Although structured, the derived program space remains large and contains heterogeneous, conditionally dependent design choices whose effects can only be assessed via downstream execution. We therefore propose MemEvo, an LLM-driven auto-research framework that uses pretrained LLM as a semantics-aware proposal model to iteratively generate and refine candidate memory programs based on accumulated experimental feedback. At runtime, a deterministic evaluation pipeline validates and evaluates each candidate, while the underlying vision-language model remains frozen throughout discovery. We finally produce a training-free, bounded-memory mechanism. Extensive experiments on StreamingBench and OVO-Bench demonstrate strong streaming video understanding performance together with substantial context and inference efficiency.

Pretrained transformers use little of their depth to follow references in context. Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little. A task-trained rank-8 LoRA at one early layer extends this computation with all model weights frozen. Qwen3-8B improves from 15.5% to 99% exact accuracy on 24-line chains; a longer-trained LoRA reaches 50 lines. Ouro-1.4B reaches 60 lines after four loops and at least 160 after eight. The LoRA starts a relay: program lines pass on their chain identity through a short range of middle layers. Frozen heads read progressively further up the chain, and removing parent-line attention stops the relay. A frozen-model measurement locates the last useful intervention layer within tolerance in three of four held-out models. Task-specific LoRAs also improve MuSiQue. Default answers therefore understate the computation accessible through a tiny edit. Code and an interactive demo are available at https://lunamos.github.io/stop-thinking-too-early/

On-policy distillation (OPD) reduces train-test state mismatch by training a student on its own generated trajectories, but weak students may visit teacher-misaligned prefixes where supervision is less representative. We introduce SAKI (Supervision Allocation with KL-constrained Interpolation), which combines a KL-constrained teacher-guided rollout with maximal coupling and reuses realized accept/correction events to route token-level supervision. Accepted positions retain sampled-token reverse-KL supervision, while correction positions receive direct supervision on the teacher's highest-probability token. Under maximal coupling, the correction probability is exactly TV(p_t, q_t), so the same trust-region radius controls rollout deviation and upper-bounds intervention and specialized-supervision frequency. We further implement an engine-resident speculative verifier that preserves the exact-q trajectory distribution and coupling semantics while improving matched-workload rollout throughput by 4.22x. Across seven mathematical reasoning benchmarks, SAKI improves the matched teacher-guided baseline in Mean@8 and Pass@8 for both 1.7B and 0.6B students. Placement controls and fixed-prefix analysis further support correction-triggered routing as a conflict-adaptive supervision signal.

Retrieval-augmented fact-checkers often receive a reliability label, such as HIGH or LOW trust, for each evidence source. These labels should adjust the model's confidence and its decision to search for more evidence, while the verdict should follow the evidence content. We introduce TrustSwap, a counterfactual test that swaps, lowers, or removes source labels while keeping every evidence text fixed, and measures its three output channels (the verdict, the confidence, and the search decision) separately. Across untrained and RL-trained models at two scales, three datasets, and two prompts, confidence and search respond to the labels as intended in 49 of 50 comparisons, yet a label change alone alters 4-23% of confident verdicts for Qwen3 models and up to 50% for an existing RL-trained fact-checker. Standard GRPO fine-tuning amplifies this shortcut at 8B in all six settings. To reduce it, we propose trust-swap augmentation (TSA), which trains GRPO on each claim with both its original and its label-swapped evidence under the same gold verdict. At 4B, TSA lowers the verdict flip rate by 7-35% (relative) in four of six settings, keeps accuracy and the intended confidence and search responses, outperforms reward-based alternatives in the main setting, and carries over to an unseen label-removal perturbation. An added consistency reward helps on the trained-on swap but not on unseen perturbations. At 8B, TSA's effect is not detectable, which makes scale the main open question.

Structural reasoning, the ability to recognize and make inferences over the relational structure between objects and concepts, is a hallmark of human cognition, yet prevailing methods often collapse relational topology into flat embeddings, cannot discover hidden structure and lack interpretability. We introduce Neural Structural Reasoner (NSR), a brain-inspired network that preserves relational structure directly in the connectivity and dynamics of coupled neuronal populations. NSR draws inspiration from three biological mechanisms: multi-layered architecture for encoding hierarchical knowledge, stable representations of entity and concepts, and path integration for input-driven state inference. At query time, NSR parallelizes computation over candidate relational structures and leverages confidence-weighted scores to perform link prediction. Across standard knowledge-graph benchmarks, NSR achieves competitive accuracy without leading on every dataset, and has lower reported training times than several neural baselines. Because reasoning is implemented through sequences of human-readable neuron activations, NSR affords native interpretability by tracking intermediate inference steps. The model further extracts latent relational hierarchies and compositional rules, demonstrating the brain-inspired architecture as an effective, efficient, and highly interpretable substrate for structural reasoning.

Joint speech-gesture synthesis must coordinate two modalities despite limited paired data. Existing approaches often lack bidirectional interaction, have limited language coverage, or simplify body and finger representations. We present DualTrack, which couples pretrained speech and motion priors on a shared 12.5 Hz timeline. Causal adapters exchange previous-packet information, while current-state fusion coordinates the streams before they separately complete sixteen-codebook packets. We evaluate 43 BEAT2 recordings in four languages, with speakers held out from joint training and validation. On the shared English/Spanish inputs, without speech or motion prefixes, DualTrack achieves lower word error rate and full-motion Fr\'echet Gesture Distance, higher beat consistency and speech naturalness than the evaluated GELINA baseline.

Language-model agents increasingly rely on persistent natural-language skills to adapt beyond their frozen model parameters. When a shared skill is repeatedly revised from a non-stationary, heterogeneous task stream, however, improvements for new tasks can overwrite procedures needed for earlier ones. In continual learning, Orthogonal Gradient Descent (OGD) addresses analogous interference by projecting a new-task gradient onto a subspace that locally preserves prior predictions. Natural-language skill revisions, however, have neither gradients nor a canonical vector space in which such a projection can be performed. We introduce \emph{Semantic-Scope Projected Evolution} (SSPE), which transfers the functional principle of gradient projection from parameter space to behavior space. SSPE treats an unconstrained skill revision as a proposed update, identifies acquired capabilities with which it may interfere, and uses the observed gains and regressions to construct a compatible revision rather than merely rejecting the update. This enables one shared skill to evolve across latent and recurring task contexts without exposing semantic domain identities to the evolution model. Across controlled synthetic streams and heterogeneous real-agent benchmarks, SSPE improves final cross-domain competence and mitigates forgetting relative to strong skill-evolution baselines. The evolved skill also retains the strongest average performance after transfer to a different executor model. These results establish semantic projection as a promising principle for stable and adaptive self evolution of language agents.

Modern agents increasingly rely on memories, tools, and execution logic, so their competence extends beyond model parameters. This shift exposes a limitation of conventional knowledge distillation, which asks how a student model imitates a teacher model. We define Agent Distillation as the persistent transfer of task-solving knowledge from a teacher agent to a student agent. Our study organizes the field by where transferred knowledge is retained: within the model, as artifacts, through the execution harness, or across substrates. This perspective separates transfer evidence from its outcome and clarifies how knowledge moves between agent components. We develop an evaluation framework that relates retention to causal contribution and deployed utility. Together, these contributions establish a foundation for the reliable, maintainable, and safe development of increasingly complex agentic systems.

Open-ended generation lacks canonical answers, making pointwise rewards difficult to calibrate for group-based reinforcement learning. Directly ranking same-query rollouts provides a more suitable relative reward signal, but existing ranking-based reward methods can incur substantial judging cost. We introduce RankBuffer, which maintains an ordered, query-specific buffer of previously judged responses as a reusable quality scale. Each rollout is first inserted into an anchor interval through an independent coarse judgment, after which only rollouts assigned to the same interval undergo local fine ranking. The resulting complete order is converted into bounded rank rewards, while boundary expansion, local refinement, and inactive-anchor pruning adapt the buffer as the policy evolves. Across four open-ended benchmarks, RankBuffer consistently outperforms all pointwise baselines. It also achieves nearly on-par performance with the strongest ranking-based reward baseline while substantially reducing judging cost. Ablations demonstrate the importance of both local fine ranking and anchor response content, while buffer analyses show that rollout-derived anchors progressively extend and refine the covered quality scale. These results establish response reuse as an effective approach to efficient relative reward construction.

A critical prerequisite of generative recommendation is designing semantic identifiers (SIDs) that are both scalable to large item sets and efficiently learnable. Existing SID learning methods fundamentally rely on Top-1 hard assignment during vector quantization. While heuristic strategies -- such as clustering-based initialization or forced post-hoc collision resolution -- can artificially inflate codebook coverage, they often disrupt end-to-end semantic alignment and fail to address the underlying optimization bottleneck: sparse gradient propagation. In standard Top-1 assignment, gradients concentrate on a narrow subset of frequently selected codewords, leaving the majority inherently under-trained and causing severe SID collisions. To overcome this limitation natively without relying on complex initialization priors, we propose FineSID, a unified quantization framework that moves beyond Top-1 assignment by enabling fine-grained gradient propagation across the entire codebook. Instead of updating only a single selected codeword, FineSID distributes learning signals to all codewords in a soft, differentiable manner. This design promotes globally balanced codebook optimization while strictly preserving semantic consistency, effectively alleviating SID collisions and stabilizing training in large, high-dimensional codebooks. Extensive experiments on multiple public benchmarks demonstrate that FineSID is robust to initialization configurations and consistently improves both codebook utilization and recommendation accuracy. Our work provides a principled, initialization-agnostic solution for semantic identifier learning, advancing the practicality of generative recommendation.

While the "Matthew Effect" and filter bubbles are widely recognized outcome-level biases in recommender systems, we reveal that Diffusion Recommender Models (DRMs) uniquely compound this issue through their generative dynamics. Rather than merely inheriting data imbalances, DRMs trigger a self-reinforcing amplification of popularity bias. We identify that this phenomenon is driven by two compounding mechanisms. First, while optimization loss is universally dominated by high-frequency items across recommenders, DRMs suffer from a unique structural prior mismatch during generation. Because the forward terminal distribution of long-tailed data deviates significantly from the standard Gaussian prior, reverse sampling trajectories inherently collapse toward high-density popular items, fundamentally suppressing niche item generation. To dismantle this self-reinforcing loop, we propose FairDiff, a plug-and-play fairness-aware diffusion framework. To overcome the popularity-dominated loss, we introduce Popularity Condition Guidance (PCG). Rather than altering the training objective, PCG acts as an inference-time distributional reweighting mechanism, mathematically reshaping the score-based gradient field to penalize high-popularity regions and guide trajectories toward niche semantics. Furthermore, we design a Semantic Calibration (SC) Module to bridge the prior mismatch, aligning the forward and reverse distributions via one-step optimal transport. Comprehensive evaluations demonstrate that FairDiff achieves state-of-the-art performance while effectively mitigating the self-reinforcing Matthew Effect, highlighting its value as a general framework for DRMs.

Machine learning engineering (MLE) agents have made substantial progress, but learning through ML experimentation remains costly in time and computation. Synthetic environments reduce these costs while introducing variations in data and experimental settings that require task-specific diagnosis. Access to diagnostic tools alone does not ensure that agents learn when to use them or how to act on their findings. We introduce ToolMLBench, a suite of executable tools for data inspection, code verification, and experiment diagnosis, together with an SFT and RL pipeline for learning their use. Diagnostic calls acquire evidence whose value depends on subsequent decisions, so final outcomes provide limited guidance on which calls to reinforce. We address this challenge with SPICE, which measures how privileged context changes the likelihood of a sampled tool action and uses this difference as a turn-level reward alongside the final outcome. We train on 80 synthetic tasks and evaluate on 25 in-domain and 10 out-of-domain tasks. Providing tool interfaces and descriptions alone yields inconsistent gains across unadapted models. With the same diagnostic interface, our training pipeline raises in-domain success from 24.8% to 52.4% for Qwen3-8B and from 35.6% to 69.2% for Qwen3.5-35B-A3B. The latter also improves from 31% to 48% out-of-domain, supporting learned diagnostic tool use on held-out sources and targets.

LLM judges are inherently subjective, often favoring different responses in pairwise comparison when neither option is objectively wrong. To study this subjectivity, we introduce JudgeProfile, a framework that dissects LLM evaluation into perception (how a judge compares two responses across specific attributes like clarity, correctness, and detail) and prioritization (how much each attribute influences the final choice). We curate SubjectiveSet, a dataset of 50,013 response pairs from 17 public data sources, evaluated by 21 LLM judges across 87 attributes. We find a hidden consensus in perception: judges frequently agree on attribute judgments even when their overall choices diverge. Building on this separation, we first characterize each judge's prioritization using attribute weights estimated from its own overall choices. These weights differ across judges even when estimated from the same attribute judgments. We then learn new weights from reference labels to adapt their decisions to a target evaluation standard. Reweighting perceived attributes improves average held-out agreement with reference labels from 66.48% to 71.97%, outperforming fine-tuning and rubric prompting. Our findings show that understanding and steering the subjectivity of LLM judges requires attention not only to what they perceive, but also to how they prioritize it.

Can a scientific agent distinguish a law it inferred from evidence from one it merely recognizes? We introduce Synthetic Universes, a controlled benchmark that pairs canonical famous worlds with matched twisted twins governed by nearby noncanonical mechanisms. We evaluate each reported law twice: by executing it on held-out continuations and transfer settings, and by independently checking whether it recovers the generating mechanism. In the current checkpoint of a pre-specified 60-cell study, 22 trials were graded and one additional run ended in infrastructure failure. Among 20 twin trials, 8 pass predictive verification while 5 recover the generator. The dissociation is bidirectional: six parsable outputs predict successfully while missing the mechanism, whereas three recover the mechanism but fail predictive rollout. Drag exhibits the first pattern (5/5 predictive pass, 1/5 mechanism recovery); Gravity exhibits the second (1/5 predictive pass, 4/5 mechanism recovery). Because matched famous controls, the corrected identifiability sweep, and the Evidence Ladder remain incomplete, we do not claim a confirmatory causal prior-conflict effect. Instead, the completed runs establish a narrower verification result: predictive adequacy and mechanism recovery are distinct scientific claims and require distinct tests.

Agents increasingly build on code written by other agents, and they reimplement rather than reuse, growing the codebases later agents must work in. To measure how well agents design libraries for other agents, we introduce LibraryDesignBench, a two-phase benchmark in which an agent implements a full-featured library from a specification that defines required capabilities and potential use cases without prescribing the design. We evaluate the library through the correctness and simplicity of programs written by three user agents from different model families. The benchmark spans 242 expert-validated programming problems across 15 library-design tasks in four languages. On eleven of the fifteen tasks, agent designers reproduce the abstractions of the human-written production library. Downstream agents adopt agent- and human-written libraries alike but underuse them, reimplementing capabilities the library already provides. Our failure analysis finds that downstream agents write extra code mainly because agent-written libraries are rigid or hard to use, not because capabilities are missing. We also experiment with giving designers more prescriptive, agent-first guidance and having them test their library with subagents; this improves downstream scores and yields simpler programs. LibraryDesignBench provides both a testbed for evaluating library-design practices for agent users and an initial design baseline that improves downstream reuse.

In branch-and-bound (B&B) for mixed-integer linear programming (MILP), branching variable selection critically impacts efficiency. Existing neural branching policies often require GPU inference, while CPU-efficient symbolic expressions lack the representational capacity for complex logic. Large Language Model (LLM)-generated code provides a flexible search space for designing lightweight branching rules with diverse algorithmic logic. To discover effective rules within LLM-based evolutionary frameworks, a core challenge arises: full B&B evaluation on real instances is prohibitively expensive, whereas offline imitation learning suffers from distribution shift. To address this, we introduce a Bi-Fidelity Evolutionary framework (BiFE). It employs low-fidelity imitation scores as a rapid pre-screener and selectively applies high-fidelity on-instance evaluation only to elite candidates, effectively balancing search efficiency with performance reliability. Experiments validate both the search efficiency of BiFE and the competitiveness of its discovered rules, which outperform the SCIP solver and other baselines on CPUs, and even surpass certain GPU-based neural policies.

Under irreversible resource depletion, an agent can spend resources to distinguish among latent fault models, or to change the system state so that the remaining models admit a common acceptable continuation--at which point further diagnosis becomes unnecessary. This distinguish-or-homogenize principle identifies a path that existing frameworks for identification, planning, and diagnosis do not make explicit: prior formulations treat the mapping from fault models to acceptable policies as a given, whereas LCPI makes it a function of the agent's own actions. We formalize this principle through Last-Chance Policy Identification (LCPI), where correctness is evaluated at the state the agent reaches rather than at the initial state. The Last Identifiable Margin (LIM) marks the feasibility boundary between distinguishing and homogenizing. For deterministic diagnostic graphs we provide the Exact-LIM recursion; for noisy finite-horizon recovery we propose Risk-Budgeted Compatibility Planning (RBCP), which searches a compatibility-aware frontier under a hard worst-case failure constraint. Across incident recovery on abstract microservice topologies and latent-damage navigation in MiniGrid, RBCP improves risk-feasible recovery while satisfying the failure budget. A sham control--cost-matched actions that preserve model incompatibility--eliminates the gain entirely, confirming that the benefit comes from changing which policies are acceptable for which models, not from extra search or additional budget.

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for improving large language models (LLMs) on various tasks, yet its sparse outcome rewards lack token-level credit assignment for intermediate steps. To address this, on-policy self-distillation (OPSD) leverages a self-teacher with privileged context to provide additional dense learning signals. However, because the self-teacher is often overconfident and imposes excessive penalties on long reasoning trajectories, OPSD frequently struggles in practice. To mitigate this, we propose self-instructing policy optimization (SIPO) with a contrastive self-teacher to provide dense credit. At each iteration, SIPO samples multiple rollouts per prompt from the current policy, scores them with environment rewards, and constructs two teacher contexts for each rollout by pairing the reference answer with mistakes made within the group. The model then re-evaluates its own responses under both contexts, using the difference between the two teacher log-probabilities as token-level feedback, so that biases shared by both contexts are expected to largely cancel. The resulting objective yields a token-level advantage for every rollout: the reward still sets the main direction of each update while the self-teacher redistributes credit across tokens. Even in groups where every rollout fails and group-relative advantages vanish, SIPO still provides a learning signal. By preserving direct optimization of the task reward while providing dense, token-level feedback, this approach bridges reinforcement learning and on-policy self-distillation. Extensive experiments across multiple reasoning and code-generation benchmarks demonstrate that SIPO outperforms both RLVR and OPSD baselines without an external teacher or additional generation.

Agent skills provide a lightweight mechanism for self-evolving agents to accumulate reusable procedural knowledge without updating model parameters. However, existing learned skill curators typically optimize curation without explicitly modeling downstream executor behavior. We show that this can cause systematic cross-executor degradation: curators trained with different executors perform best when paired with their own training executor, indicating that effective skill curation is executor-dependent. We formulate behavior-adaptive skill curation and introduce EASE, a framework that learns a single curator that adapts its decisions to different executor behaviors. EASE maintains an online behavioral profile of recent execution patterns and conditions the curator on this profile, the current trajectory, and retrieved skills to add, modify, or remove skills from an evolving repository. We train the shared curator jointly across multiple frozen executors with reinforcement learning, using retrieval-aware and behavior-aware temporal attribution to focus optimization on curation actions with observable downstream influence. Across ALFWorld, ScienceWorld, and WebShop, with executors ranging from Qwen3-8B/32B and GPT-OSS-120B to unseen Kimi K2.6, DeepSeek V4 Flash, and Gemini 3.5 Flash, EASE outperforms strong skill- and memory-based baselines without per-executor finetuning. EASE also maintains 34.5--41.0% fewer skills, improves skill retrieval by 36.3--38.7% and measured edit utility by 51.8--60.0%, and reduces deployment-time inference tokens by 9.1--14.5%. These results establish behavior-adaptive skill curation as an effective principle for building self-evolving agents.

Knowledge editing enables rapid updates of specific factual knowledge in large language models (LLMs) without full retraining. However, more realistic scenarios call for a lifelong framework that handles continual updates rather than one-off modifications. In such settings, existing editing methods often overfit to target prompts, significantly degrading both the generalization of the edited knowledge and the model's general capabilities. To address this issue, we propose GLIME (Generalizable Lifelong Model Editing), which combines knowledge editing with preference optimization over generation behavior. GLIME further incorporates replay-based editing and a gradient constraint to preserve previously edited knowledge. Experimental results show that GLIME significantly improves knowledge generalization in lifelong editing settings while maintaining both editing performance and general capabilities.

Frontier coding agents can now write and execute code that authors 3D environments, but whether they reliably understand 3D structure and precisely control scene state remains unclear. The generated 3D scene is a persistent, executable artifact: a convincing render can hide incorrect spatial relations, intersecting objects, or unintended modifications. We introduce Code4Scene, a benchmark of 190 Unreal Engine cases built from human-assembled scenes that evaluates coding agents on two complementary settings under a shared execution interface. Construction tests scene-level spatial reasoning from open-ended language specifications, where many realizations are valid; editing tests precise control of scene state, where the agent must recover the target scene from reference images while preserving everything else. Rather than scoring code or rendered views, Code4Scene evaluates the generated engine-native scene for task fulfillment, artifact integrity, and static physical validity, with edits additionally compared against withheld ground truth. Across 14 coding-agent configurations on the 95-case public set, construction and editing performance are strongly correlated but not interchangeable (Spearman $\rho = 0.78$): Claude Fable 5.1 leads construction, Gemini 3.8 Flash leads editing, and GPT-6 Astra narrowly leads overall. Spatial Composition is the weakest construction category for every agent, while editing remains imprecise: the best Repair F1 is only 0.527, and 35.8% of edits that fully recover the target still introduce unintended changes elsewhere in the scene. These results expose a gap between plausible 3D generation and reliable spatial reasoning and state control.

Token compression combines content from several positions, yet rotary position embeddings usually assign the merged token one coordinate. We ask what positional information must survive later merges. Aperture stores Fourier moments of the token's weighted support at the model's rotary frequencies. We prove that these moments have minimal real dimension among continuous states sufficient for the selected expected rotary interactions. Represented mass makes updates additive; attention normalisation remains a separate readout choice. Uniform intervals give a centre rotation times a sinc gain. We characterise when centres determine interval widths and construct matched examples where they do not. Numerical checks verify the weighted-support implementation. In trained temporal readers, compression transfer varies with gain calibration and feature placement. In a prespecified native video question-answering comparison, stored support reaches $65.63\%$ accuracy versus $67.12\%$ for the deployed merging rule. These results separate exact positional preservation under compression from downstream benefit.

Compiler backends are expensive to build and maintain as programming models, workloads, and accelerators evolve. We investigate whether large language models can replace the conventional optimizing and lowering pipeline, a process that we call AI lowering. We study AI lowering from Triton to NVIDIA PTX: an LLM agent translates Triton kernels directly into PTX. We build an environment that evaluates candidate PTX, and an agentic harness in which an LLM translates Triton kernels into PTX. Across twelve common kernels on Ada, Hopper, and Blackwell GPUs and ten kernels from recent ML papers, AI lowering achieves 0.83x-3.34x the performance of autotuned Triton. The largest gains come from transformations that Triton's lowering pipeline does not perform, such as decoding packed binary weights directly into Tensor Core operands (3.34x on BitDelta), assigning each thread a complete softmax row in tensor memory (1.37x on FlashAttention), and reusing overlapping convolution windows (up to 2.23x). These results rely on a robust evaluation harness with comprehensive verification support. We build on Volta, an existing PTX verifier, and substantially extend it to support modern GPU architectures by introducing support for Blackwell's tcgen05 Tensor Core interface. This requires modeling three architectural features: managed tensor memory, descriptor-based operand layouts, and asynchronous execution coordinated through commits, waits, memory barriers, and proxy fences. We discuss the challenges involved in formalizing them, as well as the current limitations. Our results suggest an emerging future in which AI compilers replace custom-written intermediate representations and checkers, reducing the time and engineering effort required to bring up software for new general-purpose and custom chips.

Large language model (LLM) agents reuse external memory to guide new tasks, but effective retrieval requires learning which memory sets improve execution. Such learning relies on costly outcome feedback: ordinary retrieval observes only executed sets, while evaluating alternatives requires additional rollouts. We introduce \textsc{UpliftMem}, which learns memory retrieval from set-level execution uplift relative to the same executor without memory. A theoretical analysis of how retrieval preferences restrict feedback coverage motivates targeted probing of alternative memory sets. Probe selection follows an expected value of sample information (EVSI) criterion, derived in closed form under a correlated Gaussian model, to allocate limited training rollouts according to their expected improvement in local retrieval decisions. The shared scorer is trained with a frozen executor and selects memory sets without test-time probes. Across ALFWorld, WebShop, and BigCodeBench, \textsc{UpliftMem} achieves the best success rates among evaluated baselines on the main evaluation sets. Controlled fixed-store and matched probe budget evaluations further demonstrate improved memory-use decisions and more effective use of execution feedback.

Modern engineering systems, from automobiles to aircraft, are designed by using precise, continuous parametric computer-aided design (CAD) models. Evaluating design changes through numerical simulation requires meshing the continuous geometry, a computationally expensive and often brittle process that can require manual intervention and replaces the continuous representation with a discrete approximation. Most neural surrogates accelerate the simulation, but inherit this representation gap by relying on meshes, point clouds, voxels, or other sampled approximations of geometry. We introduce CANTO, a transformer neural operator that maps directly from continuous CAD geometry to physical fields, without meshing the input geometry. We develop a theoretical framework for learning operators from geometric manifolds to function spaces of physical fields, representing geometry through sequences of parametric patches. CANTO instantiates this framework by directly tokenizing non-uniform rational B-spline (NURBS) patches from their control points, knot vectors, and weights, and predicts continuous surface and volume fields at arbitrary query locations. We evaluate CANTO on four automotive and aircraft aerodynamics industry benchmarks: AhmedML, WindsorML, DrivAerML, and HiLiftAeroML. CANTO achieves state-of-the-art accuracy on most evaluated surface and volume prediction tasks, including a 19.8% reduction in surface-pressure relative $L_2$ error compared with AB-UPT on HiLiftAeroML. Differentiability with respect to CAD parameters further enables gradient-based inverse design of designs. On AhmedML, CANTO identifies designs with 4.4 to 20.4% lower drag than the best dataset designs satisfying the same volume and lift constraints, with the improvements verified using the same CFD setup used to generate the original dataset.

Embedding-based retrieval is attractive for long sequence collections because each item can be encoded once and searched by nearest-neighbor ranking. The difficulty is that the objects being indexed are often observed under a noncanonical clock: cardiac cycles stretch with rate, speech changes with tempo, and sensor traces reach comparable states at different speeds. This paper studies a specific source of instability in patch-based encoders for this regime. If patch boundaries are chosen from signal geometry, then the tokenization can change under the same temporal deformation that the representation is expected to tolerate. We propose GeoPatch, a fixed-scaffold patch encoder that keeps token support independent of geometry and uses slope, curvature, acceleration, affine-residual, and confidence descriptors only as continuous conditioning variables. The design turns boundary variation into feature modulation: geometry can change the embedding through a controlled pathway, but it cannot change the number, order, or support of local tokens. We formalize this distinction through a mechanism-level stability analysis that separates boundary drift, affine timing variation, confidence-weighted geometry perturbation, and retrieval-margin effects. The same local tokens support global embedding retrieval and late-interaction scoring, so the scoring rule can be matched to the evaluation protocol. Across ECG, speech, and multivariate time-series retrieval tasks, GeoPatch improves early-rank retrieval under timing variation while exposing a clear trade-off between local surface matching and strict non-overlap retrieval.

Post-training quantization (PTQ) lowers deployment cost for multimodal large language models, but calibration typically reconstructs fixed sequences with local objectives. This overlooks autoregressive feedback: a quantization-induced token change redirects the prefix and changes future states. Yet on-policy coverage alone is insufficient because many decision mismatches barely affect future generation. We propose OnPTQ, an on-policy framework that calibrates on trajectories visited by the current quantized policy. On shared prefixes, OnPTQ identifies quantization-eroded boundaries, evaluates competing tokens through short counterfactual rollouts, and combines current discrepancy with branch consequence into a Decision--Consequence risk. The risk prioritizes critical states, while context anchoring and trajectory refresh preserve multimodal behavior and keep calibration aligned with the updated policy. We further derive a Decision--Consequence bound linking behavioral deviation to current policy discrepancy and action-conditioned future-value span. Across vision--language and omni-modal Qwen models under multiple low-bit settings, OnPTQ improves downstream performance and yields fewer correctness flips against the corresponding Dense/FP16 references, without changing the deployed inference graph.

An executor can respond strongly to a change in a supplied plan's priority while showing a small change in the same information-selection probability when a default-aligned whole plan is removed. We call the risk of interpreting the latter as weak responsiveness to alternative priorities the default trap. We compare paired plans that prioritize different information targets with a shared no-plan reference. An accounting identity relates these distinct behavioral contrasts. Across 3,200 decision windows on 160 selected Retail, Airline, and AgentDojo tasks, switching priorities strongly redirects two models' choices, while the two plan-versus-default contrasts differ. In 2,160 additional windows, reversing account-list order shifts default target selection by 63.3-98.3 percentage points; priority-switching effects remain 96.7-100.0 points in either order. A separate 3,240-window component study finds strong control under single priority sentences, with effects of additional text varying by group and direction. Finally, 1,080 full-task episodes yield observed success differences of -19.4 to +8.3 points relative to no plan. All Retail and Airline success intervals include zero; AgentDojo results describe four fixed application worlds. These findings support joint reporting of priority responsiveness, presentation-dependent defaults, and task success and cost.

Fully asynchronous reinforcement learning (RL) improves resource utilization in large language model post-training by overlapping rollout generation with policy optimization, but it also introduces policy lag as trajectories are generated and queued while the trainer continues to update. We study how this lag accumulates over a trajectory's lifetime and how it can be controlled without sacrificing the wall-clock benefits of asynchronous execution. We decompose trajectory staleness into Generation Staleness, accumulated before rollout completion, and Waiting Staleness, accumulated after a completed trajectory enters the pool. Motivated by this decomposition, we introduce PACE (Pool-Aware Control of Effective Staleness). PACE converts excess pool occupancy into an adaptive rejection budget and ranks completed trajectories using an effective-staleness score that combines Waiting Staleness with prefix-aware Generation Staleness. This avoids penalizing long or interrupted rollouts solely because they span multiple policy versions. In single-turn mathematical reasoning, PACE improves the six-benchmark average validation accuracy by 18.7\% over unfiltered asynchronous RL at the same wall-clock budget and matches synchronous RL performance with 47.1\% less GPU time. PACE also improves validation performance in multi-turn tool-integrated reasoning, outperforming both synchronous and unfiltered asynchronous RL. Further experiments with the mixture-of-experts model and an alternative RL algorithm support its applicability across model architectures and training algorithms.

Long-context large language model inference is bottlenecked by KV caches that grow linearly with sequence length. This burden is especially severe for long, reusable context prefixes, whose cache must serve many downstream queries. Reconstruction-based methods such as Attention Matching achieve strong downstream task performance with compact KV caches. However, iterative anchor search dominates the compaction cost of OMP-based Attention Matching. This motivates our selective amortization principle of learning a reusable anchor-selection policy across contexts while retaining context-specific reconstruction. In this work, we propose ARC-KV, a novel reconstruction-based KV cache compaction method that follows this principle. To this end, we first train a value-aware indexer to select real-key anchors in a single scoring pass. ARC-KV then applies convex-hull-constrained key merging and fits an attention-mass bias and compact values against the full cache. At inference time, ARC-KV builds the compact cache once per context using the frozen indexer and reuses it for all subsequent queries. Extensive experiments demonstrate that ARC-KV outperforms reported compaction methods in most settings across QuALITY, RULER, and LongBench on Llama-3.1-8B-Instruct. In particular, at 10% KV retention on QuALITY, ARC-KV improves accuracy from 0.6409 to 0.6474 over Attention Matching while reducing compaction time by a factor of 25.73, from 959.8 s to 37.3 s.

Percutaneous iliosacral screw fixation is an important minimally invasive treatment for unstable pelvic fractures. Because the sacroiliac region has complex anatomy and narrow screw corridors, the accuracy and safety of screw placement directly affect surgical outcomes. Accurate and reliable preoperative screw planning is therefore essential to improve surgical success and reduce intraoperative risks. Conventional preoperative planning typically requires surgeons to determine screw trajectories through manual measurements, a labor-intensive process that depends on subjective clinical experience. To address these challenges, we propose a fully automated pipeline for preoperative iliosacral screw planning in patients with pelvic fractures. Using patient-specific three-dimensional anatomy, the pipeline automatically identifies safe screw corridors and generates individualized insertion trajectories to support clinical preoperative planning. We evaluated the proposed pipeline on 200 clinical cases of pelvic fractures. Compared with conventional manual measurements, the safety margin of the safe insertion corridors increased by 2% across the four screw types, the mean planning time decreased by more than 90%, and the clinical acceptance rate reached 95%.

Multi-agent LLM systems rely on message passing among specialized agents to accomplish complex tasks. However, an upstream agent may provide useful information or an incorrect answer that causes a downstream agent to override a correct answer supported by its own evidence. Prior work has not clearly separated the benefits of communication from the damage caused by incorrect messages. We study this problem with controlled experiments across five benchmarks and five receivers, keeping the downstream task and evidence fixed while comparing answers under three conditions: no message, the upstream agent's original message, or a message with the opposite conclusion. Our experiments reveal three key findings. First, messages often help when the downstream agent would otherwise answer incorrectly. Second, messages can also hurt: when the downstream agent would answer correctly without a message, an incorrect upstream message changes the answer in up to 32% of cases. Third, in 94% of audited harmful cases, the downstream agent copies the upstream's specific wrong answer--a pattern we term answer substitution. Removing unreliable messages recovers part of the lost accuracy, suggesting that communication should be selective based on upstream reliability and the evidence already available to the downstream agent.

We present IronLLM-0.6B, a 654M-parameter language model designed for efficient on-device inference. IronLLM-0.6B combines a hybrid attention architecture with X-MTP, a lightweight shared-KV multi-token prediction design that eliminates per-depth KV-cache replay and employs a lightweight verification head for rollback-free drafting, achieving a 1.48x decoding speedup. The model is pretrained on approximately 6.2 trillion tokens using a quality-oriented data pipeline and is further post-trained with Multi-Domain On-Policy Distillation to integrate capabilities from domain-specialized teachers. To better meet the low-latency requirements of on-device scenarios, IronLLM-0.6B adopts an Instruct-Only design. Evaluations show that IronLLM-0.6B achieves competitive performance relative to larger models such as Qwen3.5-0.8B and MiniCPM5-1B, while producing more concise responses on many tasks. We further present IronLLM-0.6B-Light, which replaces RMSNorm with Dynamic Tanh and simplifies several computationally expensive components to improve inference and quantization efficiency. Together, the IronLLM models provide an effective performance-efficiency trade-off for resource-constrained deployment.

We propose STRAT, an auxiliary task that trains deep reinforcement learning (RL) agents to predict a short textual trace of their own state. Inspired by human spatial navigation, the description combines landmark, route, and survey knowledge, tracking the agent's position, inventory, goals, and immediate progress. Environment rules generate this text online without human labelling. Our method adds a single auxiliary head to a standard policy. Across 60 sparse-reward XLand-MiniGrid tasks, STRAT solves complex environments where standard RL fails outright, while compacting state representations and preventing rank collapse. Beyond performance gains, the predicted trace provides a readable account of agent beliefs at every step for no extra cost.

Recent efforts to scale tool-use post-training have largely centered on the synthesis of executable environments, which constitute only one component of a broader agentic interaction system comprising the environment, task, agent harness, and evaluator. Scaling environments in isolation, however, does not guarantee commensurate gains in model performance, because reliable learning signals depend on coherent interactions among all components of the agentic interaction system. To address this problem, we introduce WEFT (Whole-system Evolution For Tool-use Post-training), which couples scalable agentic interaction system construction, execution-driven self-evolution, and stable post-training. WEFT scales agentic interaction system construction across environment breadth, task complexity, and interaction diversity. Execution-driven self-evolution iteratively uses execution traces and state evidence to attribute failures and revise the responsible components, with fresh rollouts evaluating the changes and providing evidence for subsequent evolution rounds. For stable post-training at scale, WEFT addresses both optimization and execution reliability: prefix-preserving sampling retains verified progress and atomic-turn credit assignment localizes learning signals, while MegaMCP maintains isolated, recoverable state across concurrent rollouts over shared tool services. Extensive experiments across various models and benchmarks demonstrate the effectiveness of WEFT for tool-use post-training. WEFT-8B and WEFT-14B outperform all evaluated matched-size environment-scaling baselines on BFCL V4, $\tau^2$-Bench, and Claw-Eval. In particular, WEFT-14B improves over Agent-World-14B by 6.41, 2.23, and 12.27 percentage points. WEFT-35B-A3B further extends these gains to more challenging long-horizon workflow benchmarks, including Toolathlon-Verified and AutomationBench.

Mixed human-LLM documents require locating authorship transitions from detector scores whose reliability varies across text units. Existing weighted mean contrasts are vulnerable to extreme scores, while directly replacing means with robust centers obscures how a misplaced boundary changes the population objective. We propose Robust Weighted Profile-Loss Change Point Detection (RWCP), which combines capped reliability weights, Huber profile gains, and narrowest-over-threshold search in reliability coordinates. Our key analysis expresses the population gap between a true and a displaced split as a merge cost, avoiding a closed-form solution for the nonlinear center of a mixed segment. Under explicit curvature, spacing, and dependence conditions, core RWCP recovers the number of changes and localizes their boundaries; its quadratic-loss limit recovers squared weighted CUSUM. We also study RWCP-R, a separately evaluated decoder that shares source centers across nonadjacent passages. Across five retrospective cached-score benchmark families, core RWCP reduces family-macro WindowDiff by 17.6\% relative to weighted change-point detection, and RWCP-R lowers it further. Boundary recovery improves most clearly for isolated changes, while both fixed configurations miss changes in collaborative and densely alternating text.

As the capabilities of large language models (LLMs) continue to advance, increasing attention is turning to how to translate their abilities into useful behavior. Personal agents bring this question into everyday settings, where models are expected to serve individual users and continually adapt to their preferences. With the underlying model held fixed, such adaptation relies on harness engineering: designing and evolving the surrounding layer that manages context, memory, tools, and execution. Despite rapid progress, the factors governing effective harness evolution remain insufficiently understood. To narrow this gap, we investigate three central questions concerning harness architecture, harness scale, and self-evolution algorithms through complementary empirical and theoretical analyses. Empirically, we introduce a preference-oriented benchmark and systematically characterize the capabilities and limitations of personal agents associated with these three dimensions. Theoretically, we formulate harness evolution as a learning problem and explain these phenomena through approximation, generalization, and optimization errors. Analyses of reachable policies, capacity under finite interaction evidence, and biased update dynamics provide theoretical accounts of the observed phenomena. Together, these results offer a unified perspective on the limits of personalization through harness evolution and inform future harness design.

Recent advances in generative planning have made trajectory inpainting a promising approach to offline goal-conditioned reinforcement learning. However, these methods typically specify the planning horizon before generating plan content, even though the appropriate horizon depends on the route itself. A horizon that is too short can force infeasible transitions, whereas one that is too long can introduce redundant motion. We introduce HorizonFlow, a hierarchical planner that treats plan length as an output of generation rather than a prescribed input. Its subgoal route planner guides its action-prefix controller through a sequence of latent subgoals. Both components combine insertion-based generation with flow matching to jointly generate continuous plan content and length, using the partially generated plan to guide token insertion. HorizonFlow reuses the resulting length information to select candidates and steer generation toward shorter plans without a separate learned value model. Across Maze2D, Multi2D, and OGBench navigation and visual manipulation benchmarks, HorizonFlow achieves the highest average performance among the compared methods.

Reward hacking occurs when policy optimization exploits a brittle reward interface or an overly permissive proxy objective, improving the training score without improving the underlying response quality. This phenomenon is amplified in group-relative policy optimization: an unsupported reward can shift the group baseline and alter the updates of other rollouts, while post-hoc or purely relative weighting cannot represent group-wide uncertainty. We propose \emph{Self-Tuned Anchored Reliability Group-Relative Policy Optimization} (STAR-GRPO), a reliability-first advantage estimator based on paired assessments of the same rollout. STAR separates the quality signal from its learning influence: score disagreement determines rollout reliability, relative reliability enters a self-tuned robust location--scale fit before group normalization, and absolute group reliability attenuates the resulting bounded advantage. The analysis establishes coordinate and second-moment bounds, characterizes exact centering through the weighted location equation, and gives reliability-dependent attenuation guarantees for outlying rewards. We evaluate STAR-GRPO in two complementary reward-hacking regimes. In token-interface exploitation, STAR prevents runaway optimization of the deployed-interface score while improving the canonical quality signal. In rubric-proxy overoptimization for medical reasoning, STAR improves independent semantic evaluation, narrows the proxy--judge discrepancy, and reduces overclaim while optimizing the same task proxy. Together, these results show that reliability-first normalization offers a principled way to limit unsupported reward influence on both group baselines and policy updates, while retaining the task reward as the optimization target.

Existing reactive Graphical User Interface (GUI) agents often fail in long-horizon, dynamic scenarios, where unexpected disturbances trigger attention-diverting and cascading failures. To address this, we propose PrecogUI, a pre-cognitive architecture that shifts the paradigm from reactive execution to proactive decision-making. Specifically, we design a Proactive Experience Pool (PEP), which caches recurring anomaly and success patterns as "state-action-result" tuples in a dual-memory repository. Furthermore, we introduce a Proactive Simulation Executor (PSE) that learns to forecast the next symbolic UI layout given a candidate action, enabling early anomaly avoidance and ranking candidate actions by predicted reliability. Finally, a Pre-cognitive Execution Controller (PEC) fuses these priors and predictions, prioritizes handling of foreseen anomalies, and ensures execution robustness through a closed-loop error correction mechanism. For robust evaluation, we develop AutoTraj, an automatic data-generation engine, to construct InterfereBench, a benchmark for long-horizon tasks with strong disturbances. Experiments demonstrate that PrecogUI surpasses state-of-the-art methods on InterfereBench while maintaining competitive performance on public benchmarks. The code will be publicly available.

Many computer tasks recur: the same workflow runs many times, with new inputs and from different starting states. Current computer-use agents re-plan every step of every run, which makes them costly and unreliable on such tasks. We introduce neuro-symbolic computer use, in which a recurring workflow is executed by a learned policy rather than re-derived by an agent on each run. The policy fixes the decisions that are stable across runs (ordering, variables, loops, and branches) in executable code, and delegates observation-dependent decisions, such as grounding and state checks, to neural models. We learn these policies with neuro-symbolic policy iteration: starting from one agent trajectory, it executes the policy, diagnoses failures with task-completion and step-level judges, and revises the code with a coding model informed by an agent's continuation from the point of failure, without access to the benchmark evaluator. Iterating on generated parameter and initial-state variants makes the policy reusable, and a pre-action verifier guards each state-mutating step at deployment. On OSWorld-Verified and ScienceBoard, the learned policies achieve the highest Pass^3 of all methods in all four settings, 3.6-15.8 points above the base agent, while cutting per-run cost by 15-217$\times$ and latency by 3.4-5.1$\times$. On OSWorld-Verified, policies built only on variants transfer to the held-out original tasks, exceeding AutoRPA by 8.6-17.5 points in Pass^3.

Recently, Group Relative Policy Optimization (GRPO) and its variants have been developed for policy optimization and demonstrated notable performance gains. However, these methods usually incur substantial computational overhead due to per-question multi-rollout sampling and repeated per-token probability evaluation across rollouts. Furthermore, low-information or highly homogeneous trajectories can degrade downstream learning signal efficiency, hindering model optimization and limiting final performance. To address these issues, we propose FastRL, a novel plug-and-play reinforcement learning framework that simultaneously improves training efficiency and the effectiveness of policy learning. Specifically, 1) We introduce an advantage-aware pruning strategy to selectively preserve high-advantage trajectories while maximizing inter-trajectory gradient diversity. 2) Then, we design an adaptive rollout sampling mechanism to dynamically adjust the sampling scale across different training stages based on historical pruning distributions, balancing exploration adequacy and computational efficiency. Experiments demonstrate that FastRL can be seamlessly integrated into GRPO, DAPO, and GSPO variants, achieving an average 2.07$\times$ training speedup on Geometry3K and GeoQA8K-R1V, along with an approximately 1.64\% improvement in average accuracy on visual reasoning benchmarks. Source codes will be available at https://github.com/Nicozwy/FastRL.

Traffic signal control (TSC) is essential for improving urban mobility and reducing congestion. Although roadside cameras are widely deployed at signalized intersections and provide rich visual observations of evolving traffic, existing TSC methods typically rely on manually engineered traffic states or separate perception modules, creating a gap between physical observations and control decisions. We present VLALight, the first vision-language-action (VLA) model for end-to-end traffic signal control from multi-view roadside videos. VLALight directly maps visual observations to coordinated signal actions through multi-target spatiotemporal traffic reasoning and topology-aware cooperative perception across intersections. To establish this capability, we develop a two-stage supervised cold-start training strategy for visual traffic understanding and signal decision-making, followed by cooperative agentic reinforcement learning that jointly optimizes local control and network-wide traffic efficiency. Furthermore, VLALight introduces adaptive fast and slow reasoning modes, enabling the policy to allocate deeper reasoning only when additional deliberation provides sufficient control benefits. Through balanced mode-aware rollouts and relative advantage optimization, VLALight learns to trade off decision quality and inference cost. Extensive experiments on seven real-world traffic-flow datasets across three urban networks demonstrate that VLALight consistently outperforms transportation-based, RL-based, and LLM/VLM-based baselines. Ablation studies validate the effectiveness of cooperative perception, network-level optimization, and adaptive reasoning. These results demonstrate the potential of VLA models for real-world physical traffic control. Our project is available at https://github.com/usail-hkust/VLALight.git.

Long-context reasoning is essential for complex and long-horizon tasks, yet the performance of large language models (LLMs) degrades as context length increases. Recent approaches address this by processing input chunk by chunk while maintaining a bounded textual memory in model context. However, premature information compression can discard critical details essential for subsequent reasoning. In this paper, we introduce Commit-on-Evidence Memory (CoEM), which learns when to convert source evidence into compact memory facts. Specifically, under a fixed context-memory budget, CoEM preserves potentially useful source excerpts verbatim in a pending set, allowing subsequent context to clarify their relevance before irreversible compression. As new context arrives, a learned policy revisits each pending excerpt and decides whether to promote it to the committed memory, retain it for further consideration, or discard it. A frozen verifier ensures proposed facts are accepted only if supported by retained excerpts and current context. To further guide effective memory management, we train this policy using reinforcement learning by combining fine-grained, step-level evidence rewards with final answer rewards. Extensive experiments demonstrate that CoEM consistently improves long-context reasoning. When evaluated on 6,400 documents long-context input, CoEM outperforms the strongest memory baseline by 10.4-11.4 F1 points on Qwen3.5-9B. Code repository: https://github.com/benmagnifico/CoEM.

GUI agents automate tasks on digital devices by grounding language instructions in visual interfaces. Existing group-relative reinforcement learning improves GUI action prediction by comparing the rewards of multiple responses sampled from the same GUI state. However, binary evaluation treats spatially different failed clicks as identical and provides no relative signal when all sampled clicks fail. To address these limitations, we propose Spatial Credit Assignment (SCA), which uses the screen coordinates of sampled clicks to refine group-relative credit. Specifically, SCA predicts each held-out response's reward from the other responses in groups containing both successes and failures, then uses the prediction residual to adjust credit. When all sampled clicks fail, SCA instead orders them by distance to the annotated target. These spatial references are used only to construct the training update; the deployed policy remains unchanged. We evaluate whether this correction improves the policy update itself by comparing its error and directional alignment with the exact return gradient in a controlled synthetic study. Across GUI grounding and offline action-prediction benchmarks, SCA improves grounding across professional domains and achieves the strongest results among reinforcement-fine-tuned models on most action-prediction metrics, with consistent gains across the reported GUI suites.

Group-relative policy optimization relies on reward-derived advantages and sequence-level likelihood weights, both of which can be sensitive to localized outliers. Extreme rewards can collapse the contrast among clean responses after group normalization, while token-level log-ratio perturbations can alter sequence weights and clipping decisions. We introduce RoVR-GSPO, a dual-channel robust optimizer that addresses these failure modes separately. Its reward channel combines robust reference estimation with bounded residual credit, while its ratio channel uses differentiable SoftRoVR aggregation to construct robust sequence weights. We provide stability and efficiency analyses for both channels. Experiments on mathematical reasoning, long-context summarization, and tool-call annotation show consistent improvements over GSPO, while controlled perturbation studies demonstrate stronger robustness to reward contamination and token-ratio anomalies.

Complex questions often require multi-hop reasoning that connects facts distributed across sources or distant regions of a long context through intermediate steps. Benchmarks commonly evaluate this ability with questions built around predefined reasoning chains, treating a correct answer as evidence that the intended composition was used. Yet answer correctness alone leaves open whether success depends on the evidence associated with each intended step: models may instead rely on memorized associations, shorter paths, or partial evidence. We examine this dependence using the Behavioral Necessity Rate (BNR), which measures how often targeted evidence removal prevents answer recovery on initially correct instances. Across five existing benchmarks, panel-mean BNR ranges from 16.6% to 48.9%, exposing a substantial gap between annotated structure and observed dependence. Guided by this diagnosis, we introduce REALHOP, a diagnose-construct-verify framework that rebinds entities, factorizes selected relations, adds complete competing paths, and places evidence at traceable locations. Structural and semantic checks precede freezing; behavioral interventions follow. On 790 paired MuSiQue questions, REALHOP raises panel-mean BNR from 27.4% to 94.4% while retaining high Full accuracy. It also yields high BNR on REALHOP-FRAMES and REALHOP-LONGBENCH. On 216 long-context questions, the matched multiple-choice spread across 16 models grows from 13.9 to 59.2 points and persists under repeated open-ended evaluation. Together, these results show that a conceptually coherent chain and a correct final answer do not by themselves establish multi-hop reasoning. Verifying that success depends on every intended hop is therefore as fundamental to multi-hop evaluation as measuring answer accuracy itself.

Federated LoRA fine-tuning enables parameter-efficient adaptation of pre-trained models without sharing private data, but suffers from two fundamental mismatches under heterogeneous client data: a structural aggregation mismatch caused by independently averaging LoRA factors, and a statistical collaboration mismatch caused by enforcing a single global adapter across divergent clients. To address these issues, we propose CF-LoRA, a clustered federated LoRA fine-tuning framework that combines decoupled factor aggregation with adaptation-aware client clustering. CF-LoRA first learns a globally shared $A$ factor while retaining personalized $B_i$ factors, then identifies clients with similar adaptation patterns based on the cosine similarity of their learned $B_i$ factors, and finally performs intra-cluster $B$-factor aggregation with a frozen $A$ factor. By decoupling LoRA factor aggregation, CF-LoRA preserves the low-rank structure and mitigates the structural aggregation mismatch, while adaptation-aware clustering promotes collaboration among clients with similar adaptation patterns and reduces negative transfer caused by statistical heterogeneity. Experiments on four language tasks and four vision datasets with RoBERTa and ViT show that CF-LoRA achieves the highest average accuracy in both modalities while communicating only one LoRA factor per optimization round.

Knowledge Graph Embedding models have been extensively used to learn representations of entities and relations in Knowledge Graphs for predicting missing links. However, the quality of the learned representations varies a lot across different areas of the graph. If previous research has loosely linked the problem to relation types or degree bias, we show that it is more widespread and it correlates with the degree imbalance of the entities in test triples. In particular, the prediction of a target entity that has a degree much smaller than the degree of the anchor entity is extremely problematic. This is critical in recommender systems and other use cases, where these triples represent important corner cases. To address this issue, we propose an inference-time latent search optimization method capable of significantly improving model predictions on the most imbalanced triples. Built on top of a pre-trained model, it explores the embedding space at evaluation time, blending known and out-of-band information to mitigate the degree imbalance bias. We show the value of our approach on imbalanced triples from common benchmark datasets, where we outperform conventional methods, opening the door to the successful adoption of Knowledge Graph Embedding models on these critical corner cases.

For a long-horizon agent, context is the bottleneck: the history is resent with every request, the window caps task length, and reasoning degrades as the history grows. Replacing structured objects with compact retrieval Cards shortens the prompt and keeps the exact originals retrievable, but editing the history can break prefix-cache reuse, and prior recoverable methods time their edits by forecasts of future reuse or by preset intervals. We propose CADOC (Cache-Aware Dynamic Object Context), an online algorithm that replaces structured objects with compact Cards while preserving exact, on-demand retrieval of their original contents. CADOC schedules replacements in batches by balancing accumulated waiting cost against shared cache-reconstruction cost. Its scheduling rule follows from an economic order quantity trade-off, recovers the optimal integer batch under stationary assumptions. Across evaluation, CADOC consistently achieves the lowest aggregate input cost among the compared configurations, which reduces input cost by approximately 40\% on average while maintaining task performance close to full context. CADOC thus provides a cost-derived approach to compressible context management, demonstrating that efficient compression depends not only on shortening prompts but also on scheduling edits to preserve cache reuse.

Efficient patient flow coordination across autonomous hospital departments is critical for mitigating overcrowding and balancing resource utilization. While classical queueing theory, specifically open Baskett--Chandy--Muntz--Palacios (BCMP) networks, provides an interpretable mathematical topology for healthcare operations, analytical models rely on stationary assumptions and fixed routing matrices that degrade under state-dependent real-world dynamics. Conversely, centralized reinforcement learning approaches struggle to accommodate the decentralized structure of hospital governance, where individual clinical departments function with localized observations, heterogeneous resources, and divergent operational objectives. In this paper, we present a Multi-Agent Systems (MAS) framework titled \emph{Physics-Informed Multi-Agent Coordination}, which embeds empirically calibrated BCMP queueing topologies as physical priors within a decentralized multi-agent reinforcement learning architecture. Formulated as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) under coupled resource constraints, our method enables autonomous departmental agents to cooperatively negotiate patient routing and dynamic service scaling. To mitigate environmental non-stationarity without inducing excessive communication overhead, agents exchange localized action fingerprints along network edges and optimize a spatially decomposed reward structure. Empirical evaluations driven by real-world MIMIC-IV patient trajectories indicate that this cooperative multi-agent approach substantially reduces cumulative system delay compared to static Markovian approximations, heuristic dispatching, and independent multi-agent baselines, while maintaining clinical safety constraints.

Integrating transcriptomic and electrophysiological data is essential for building multimodal foundation models for neuroscience. Patch-seq provides paired measurements of gene expression and intrinsic electrophysiology from the same neuron, establishing a basis for training cross-modal models. Here we introduce LangPatch, a foundation-model-based contrastive learning framework that uses paired Patch-seq data to align pretrained GenePT representations with electrophysiological phenotypes through a language-based interface. Gene descriptions and verbalized electrophysiological profiles are embedded by the same frozen text encoder. A context adapter and projection modules connect the modalities through paired contrastive learning. Across mouse visual, mouse motor, and human cortical cohorts, LangPatch achieves the highest mean transcriptome-to-electrophysiology prediction correlation among the evaluated foundation-model and representation-learning methods. It also improves held-out cross-modal alignment in the two mouse cohorts (FOSCTTM 0.107/0.135 vs. 0.208/0.222 for JAMIE, an existing cross-modal Patch-seq imputation method). It predicts transcriptomic family, type, cortical layer, and marker-gene expression from electrophysiology, exceeding other baselines on most endpoints. More importantly, the method transfers across brain areas and species: a model trained on mouse visual cortex predicts electrophysiology in motor cortex with approximately 70% correlation retention and in human cortex with 47% (58% on acute-slice recordings). Together, these results demonstrate alignment between molecular and functional representations of neurons, providing a building block for multimodal foundation models in neuroscience.

As large language models (LLMs) advance, AI agents are increasingly deployed in open-world environments to tackle complex sequential tasks (e.g., document processing, cross-application collaboration), relying heavily on actions ranging from GUI operations to semantic APIs. However, three core challenges persist: the "scale dilemma" of massive tool ecosystems exceeding LLM context windows, the "non-stationarity" of tool quality due to updates or outages, and the "heterogeneity" of feedback formats (pixels, text, structured data) creating information silos. To address these, we propose AnyAct, a universal action layer that unifies available capabilities into a self-evolving action space, enabling agents to operate efficiently and reliably in large-scale, dynamic tool ecosystems. AnyAct's core design focuses on two objectives: constructing this action space via hierarchical progressive retrieval (filtering task-relevant actions) and test-time reliability evolution (pruning unreliable actions), and enabling reliability-aware action orchestration through a heterogeneous observation grounding module that unifies multi-modal feedback. Additionally, it defines a hybrid action space (primitive + semantic actions) and optimizes for a balance between task success rate and execution cost. Evaluations on LiveMCPBench and OSMCP (a new benchmark we developed for multi-granularity action collaboration) demonstrate state-of-the-art performance. AnyAct delivers substantial performance gains over baseline methods across various LLM base models on LiveMCPBench and improvements are particularly notable for models with constrained native capabilities. On OSMCP, it achieves 77.27% overall success with only 50 steps, which is half the steps required by most competitors.

Low-Rank Adaptation (LoRA) is a widely used approach to parameter-efficient fine-tuning (PEFT), yet a performance gap can remain relative to full fine-tuning (FFT). Many LoRA variants improve the initialization or optimization of low-rank factors. At each training step, however, their first-order weight-space directions are constrained by the current parameterization. We characterize the corresponding LoRA-accessible gradient space and show that it coincides with the tangent space induced by the current LoRA parameterization. This characterization yields an orthogonal decomposition of the full weight gradient at the current model parameters. We term the component orthogonal to this space the normal gradient. Based on this decomposition, we propose GDLoRA (Gradient-Decomposed Low-Rank Adaptation). GDLoRA reconstructs the full weight gradient from forward activations and backward signals, extracts its normal component, and directly updates the base weights with this component, while retaining standard AdamW optimization for the LoRA factors. GDLoRA incorporates complementary normal gradients without increasing standard LoRA's optimizer-state memory budget under matched adapter and optimizer configurations. Experiments on natural language understanding, mathematical reasoning, commonsense reasoning, and image classification show that GDLoRA consistently improves over LoRA and narrows the performance gap to FFT. The code is available at https://anonymous.4open.science/r/GDLoRA.

Federated parameter-efficient fine-tuning enables clients to adapt pre-trained models without sharing raw data or communicating the full model, but statistical heterogeneity makes a single global adapter insufficient for personalized prediction. Existing personalized methods typically use the same low-rank structure for both shared and private adaptation, overlooking their distinct requirements for aggregation and personalization. We propose FedLAFP, a role-aware framework that couples a compact, globally aggregated LoRA branch with a client-private, full-rank-capable RandLoRA branch. The shared branch provides an efficient interface for transferring common knowledge, whereas the private branch combines fixed random low-rank bases with learned scaling coefficients to provide expressive client-specific adaptation without additional communication. Client- and layer-specific mixing coefficients jointly fuse the two branches, and only the shared LoRA parameters are exchanged. A controlled linear study supports this role assignment: LoRA yields more aligned client updates and lower aggregation error, while RandLoRA more accurately recovers client-specific residuals. Experiments across four visual recognition benchmarks show that FedLAFP consistently outperforms local-only and federated LoRA baselines, achieving an average personalized accuracy of $86.93\%$ and exceeding the best baseline average by $1.30$ percentage points.

Streaming video assistance requires models to answer asynchronous questions from an observed prefix under a fixed context budget. Existing approaches model response timing or compress history, but an online state formed before future questions are known can omit visual details before later questions reveal their relevance; the retained state alone cannot recover them. We introduce Watch-Think-Interact (WTI), a closed-loop framework for multi-question streaming video reasoning. WTI maintains compact natural-language memory entries tagged with source-video time ranges; these entries support direct reasoning when sufficient and otherwise anchor selective recall of finer visual evidence. For each question, WTI answers when current context and memory suffice, continues watching when required evidence has not appeared, or recalls a relevant past interval and decides again after incorporating the returned chunks, without replaying the full observed history. To train this behavior, we construct WTI-82K, comprising 82,335 timed questions across 4,812 causally aligned trajectories, and develop Stream-GDPO to optimize complete multi-question streaming rollouts using trajectory-level feedback for response timing, source-video recall, and memory updates. WTI achieves state-of-the-art aggregate performance among the compared open-source streaming baselines, reaching 83.3% on StreamingBench and 73.6% weighted overall accuracy on OVO-Bench.

Multimodal GUI agents have achieved impressive results on general software benchmarks, yet their ability to operate professional scientific software remains largely unexplored. In materials science, sparse domain-specific web data, specialized interfaces, and tacit workflow conventions create blind spots that general-purpose pretraining cannot readily bridge. We present MatToolBench, the first real-environment benchmark for evaluating multimodal GUI agents on professional materials science software, comprising 204 tasks across 10 tools in three modalities: GUI operation, OriginPro scripting, and code-based database queries, all executed inside a Windows 11 VM. Each task is decomposed into fine-grained sub-criteria by domain experts, enabling interpretable partial-credit scoring; the GUI component of our multi-level evaluation pipeline achieves an average F1 of 0.98. For OriginPro figure-generation tasks, we further conduct a human-LLM agreement study to validate the use of a multimodal judge for secondary aesthetic assessment. Our experiments show that strong performance on general benchmarks does not transfer to professional scientific workflows, and that this gap is not a visual-grounding problem alone: failures arise from domain-specific operational knowledge, sparse pretraining coverage of scientific software, weak cross-tool artifact handoff, and critical states exposed only visually. Even the best model reaches only 25% success rate on GUI tasks and 45% on code tasks. MatToolBench therefore serves as a challenging diagnostic benchmark and real-environment testbed for data-scarce, knowledge-intensive scientific workflows.

Ensuring the safety of reasoning large language models (LLMs) across languages is essential for their reliable deployment. However, when exposed to jailbreak attacks in non-high-resource languages, these models may generate unsafe responses even when their reasoning traces identify safety risks. To address this issue, we propose aligning cross-lingual thoughts and responses (ACTR), a framework that improves multilingual safety alignment by strengthening the use of existing safety reasoning. Specifically, we first present the think gap score (TGS) to compare the normalized contributions of reasoning traces to attention outputs during response generation across languages, and use reasoning- trace substitution to measure the cross-lingual safety gap. Next, using a corpus of jailbreak queries, we assess neuron importance through changes in response representations caused by neuron masking and compare the high-importance neuron sets obtained with reasoning enabled and disabled to identify safety think neurons that support the use of safety reasoning. Finally, we devise neuron-selective consistency optimization (NSCO), which uses a frozen judge model to reward agreement between the safety categories of reasoning traces and responses while updating only the parameters associated with the selected neurons, requiring no human-annotated responses or preference data. Across two reasoning models, ACTR achieves lower average attack success rates than the evaluated state-of-the-art methods on AdvBench-X and MultiJail, with safety gains extending to unseen languages, while preserving or improving average performance on multilingual knowledge and mathematical reasoning tasks and limiting false refusals of benign requests. Warning: this paper contains examples with unsafe content.

The rapid growth of submissions and reviewing workload has accelerated the use of large language models (LLMs) in peer review. Prior studies suggest that LLM-based reviewers can penalize content perturbations, such as overclaiming, indicating a certain degree of reliability. Yet these conclusions are largely based on a narrow set of perturbation strategies instantiated with static templates, providing limited evidence of actual reliability. In this paper, we construct a three-level evaluation framework covering perturbations to surface presentation, argumentative logic, and value judgment. Experiments on representative LLM-based reviewers reveal two limitations of static evaluation: stratified vulnerability, where perturbation effects depend on whether the paper's original review score is high or low, and perturbation undercoverage, where a single template misses vulnerabilities exposed by diverse realizations. To address these limitations, we propose SCOPE-Fuzzer, a strategy-aware fuzzer that combines feedback-driven strategy selection with adaptive mutation of paper content. By iteratively probing reviewers with dynamic perturbations, SCOPE-Fuzzer consistently uncovers vulnerabilities overlooked by static evaluation and other baselines.

Long-horizon LLM agents require reinforcement learning methods that can assign credit to intermediate decisions under sparse and delayed rewards. Existing group-based methods such as GRPO and GiGPO alleviate this issue by comparing rollout returns or repeated anchor states, but they still fail when the compared returns have no variation. We identify this failure mode as zero-credit failure: during early training, many failed rollouts contain useful prefixes, yet existing methods assign them no task-discriminative advantage. To address this issue, we propose Milestone Viability Potential Policy Optimization (MVPO), a potential-routed policy optimization algorithm that learns from viable failure prefixes. MVPO estimates prefix potential over Union-Find viability regions, repairs zero-credit groups with potential-difference advantages, and attenuates the potential branch according to relative performance progress. Experiments with Qwen2.5-1.5B-Instruct show that MVPO outperforms eight strong baselines, including GRPO and GiGPO. Under the same training length, MVPO improves over the GiGPO baseline by +4.4 success points on ALFWorld and +5.3 on WebShop, while adding only 0.16%-0.20% advantage-construction overhead.

Prospective memory allows an agent to retain an intention tied to a future condition, but the stored intention does not reveal whether that condition currently holds. Checking it may require web access, multi-step tool use, and paid calls. Existing systems decide when intentions require attention, but do not allocate the resulting observations under a shared budget. We introduce the first resource-allocation formulation for the external observations required by stored intentions under a shared episode budget. BudgetPM offers two policy variants that share a hard-budget executor. BudgetPM-Static uses a lightweight Logistic scorer to learn whether a check improves the current decision. BudgetPM-Sequential distills full-episode hindsight schedules into a lightweight policy that decides when to spend or reserve capacity using only pre-query information at deployment. We evaluate BudgetPM against two public memory-agent systems, five matched controls, and four hand-designed monitoring or budget-adaptation rules. Across two benchmarks and three backbones, BudgetPM-Static outperforms adapted Mem0 and PMA workflows. On PM-Bench, its Logistic scorer reaches competitive quality--cost operating points alongside higher-capacity scorers and retains 99.9--100\% of unconstrained quality with 42--54\% fewer observations. Under severe scarcity and the same hard caps, BudgetPM-Sequential exceeds the strongest tested natural monitoring schedule by 1.92--2.58 Set F1 points. It reaches the same Set F1 and on-time recall with 16--33\% fewer observations. Matched attribution, exact-cost analysis, and a fixed-budget load intervention link this gain to competition between present and future opportunities. These results yield a demand--capacity design rule: local gating works when capacity covers demand, while future-aware supervision adds value when observations compete across time.

Skill evolution improves the capabilities of large language models by analyzing trajectories generated under a given skill and modifying the skill accordingly. Existing approaches typically generate a single trajectory per question. However, this provides insufficient optimization signals since it requires inferring effective skill edits from a solitary path. It is difficult to pinpoint which actions caused the failure in a failed trajectory, or to determine which actions in a successful one should be incorporated into the skill. Furthermore, they rely on a local batch of trajectories for analysis, making the optimization direction susceptible to noisy evidence. To address these, we propose SkillCome, a Skill-evolution method based on group Contrast optimization with dual memory. For each question, SkillCome generates trajectories and performs group contrast analysis to precisely identify key behavioral divergences between successful and failed trajectories, offering reliable optimization signals. The dual memory system further accumulates evidence from historical steps to track patterns shared across different groups, leading to more generalized optimization directions. Together, SkillCome builds a systematic optimization process that transforms experience from observed successful trajectories into reusable skills. Extensive experiments on six benchmarks spanning question answering, reasoning, and agentic tasks demonstrate the effectiveness of our method. SkillCome consistently outperforms baselines across five models of varying families and scales, with gains up to +5.69 points.

On-policy self-distillation (OPSD) improves large language models by letting a self-teacher with privileged information provide dense token-level supervision on the model's own trajectories. Yet existing methods typically construct the self-teacher from the current, initial, or slowly averaged policy state, leaving the quality of supervision constrained by the teacher's ability to exploit privileged information. We ask whether the model's own optimization progress can instead be recycled into a stronger self-teacher. In this paper, we introduce Bootstrapped On-Policy Self-Distillation (B-OPSD), which temporarily trains the policy ahead to obtain a future teacher, restores the student to the original policy state, and then uses the future teacher to supervise the restarted student. The future teacher improves supervision in two complementary ways, it can generate more reliable privileged trajectories and, conditioned on them, provide more informative token-level targets along the restarted student's on-policy trajectories. Experiments on mathematical reasoning with Qwen3-4B and Qwen3-8B show consistent improvements over standard OPSD in both settings, including gains from 27.50 to 41.30 and from 48.80 to 64.44 in the rollout-privileged setting. Our findings point to a broader principle for self-improving models that future learning progress can be distilled backward, preserving acquired knowledge while bootstrapping beyond the optimization state that produced it.

LLM-as-a-Judge is increasingly used to evaluate policy responses on open-ended tasks that lack ground-truth answers. Existing work often directly converts the resulting judgments into reward signals for policy training, paying limited attention to intrinsic judgment quality and largely restricting the use of Judges to training-time supervision. We systematically investigate judgment quality and downstream utility by examining both how judgments are elicited and how they are used. For judgment elicitation, we vary the Judge protocol along three dimensions: verdict granularity, critique usage, and evaluation batching. For judgment usage, beyond policy training, we extend Judge to test-time inference through Best-of-N selection, Judge-guided revision, and beam search. We find that, (i) Surprisingly, judgment quality and downstream utility do not always align. (ii) Judge protocol design substantially affects both intrinsic judgment quality and downstream utility. (iii) Judge guidance effectively converts test-time compute into performance gains, with benefits varying across inference strategies. Our results call for a multifaceted evaluation of LLM Judges on open-ended tasks, encompassing intrinsic judgment quality, and downstream utility.

Tool-augmented data agents rely on tool outputs for analytical decisions. Yet successful execution can return plausible but incorrect evidence, requiring agents to decide whether to trust or verify it. Understanding this failure requires examining both the evidence obtained through checking and the answer ultimately adopted. We introduce ToxicBench to measure checking and adoption under numerical, label, schema, and retrieval errors, pairing clean and poisoned observations over fixed source data. In the 118-task GPT evaluation across three adapters, poisoning lowers task success by 26 to 39 percentage points. Ordinary retries help under one-shot poisoning, whereas repeated poisoning reveals wrong-answer adoption after checking. Controls on three public tables isolate how supplied evidence affects recovery. After freezing the scorer, we compare its judgments with human annotations on 200 trajectories, finding 96% task-success agreement. Human judgments support retry gains over Base and confirm adoption after checking on audited tasks. We release trajectories, versioned scoring, and reference and delivery audits. These findings highlight evidence availability and answer selection as complementary dimensions of agent reliability.

Learner simulation aims to reproduce how a particular learner behaves on new tasks. Although Large Language Models (LLMs) can generate increasingly fine-grained learning behaviors, existing approaches often need to repeatedly process a growing interaction history to reconstruct the learner. This introduces additional context and inference costs and makes the acquired learner-specific simulation capability difficult to reuse across different LLMs. We therefore propose Learner2Skill, which externalizes the simulation capability acquired from historical interactions into a persistent and reusable Simulation Skill. The Skill captures the learner's current learning state and recurring response patterns, evolves as new real interactions arrive, and can be adapted to a new LLM through lightweight executor calibration without reconstructing the learner from scratch. Experiments show that Learner2Skill more faithfully reproduces fine-grained learner behavior while reducing overall token cost, and that the same constructed Skills can be effectively reused across different LLM executors.

Large language models (LLMs) have emerged as powerful tools for automated heuristic design (AHD), enabling iterative generation and refinement of heuristics. However, the dominant paradigm embeds LLMs as narrow, fixed components, such as crossover or mutation, within heavily hand-engineered evolutionary frameworks. We argue this misapprehends LLMs. It treats them as specialized tools rather than general reasoners, constrains them to low-level operations, and underutilizes their autonomy. Moreover, the extensive human priors in these frameworks violate the bitter lesson principle that general methods scaling with computation surpass hand-crafted solutions. This raises a key question: which AHD framework designs best convert stronger LLM capabilities into better heuristics? To address this, we propose metrics for LLM-driven AHD framework handcraftedness (AHI) and intelligence conversion efficiency (ICE). Evaluating ten LLMs across three challenging combinatorial optimization problems, we obtain a notable finding that frameworks with fewer human priors consistently yield higher ICE. Based on this finding, we propose SimpleEvol, an agent-loop framework for AHD which removes nearly all human priors and allows the LLM to operate autonomously. SimpleEvol consistently achieves the highest ICE, often by a large margin. Our results challenge the trend toward complex AHD pipelines and point to a lighter and more model-centric alternative, suggesting that reducing human priors is a more effective strategy to scale up with model intelligence. The source code is available at https://github.com/HenryZhu1029/SimpleEvol-Master.

Computer-use agents have become increasingly capable of executing tasks on live desktops through natural-language instructions, based on trajectories of screenshots, actions, and reasoning. We discover that they can stealthily exhibit inertia, in which they repeat fruitless actions despite recognizing that these actions are ineffective. We hypothesize that inertia is reflected in the agent's internal state, i.e., the activation values of the agent's underlying model, and propose a protocol to measure the relationship between the two. Extensive analysis of high-dimensional activation states shows that inertia corresponds to an absorbing region of activation space, where activation values become stale across actions and even after attempts to steer them. We conjecture that drastically changing the agents' activations by re-initializing them is necessary to escape inertia. Specifically, we propose R$^3$ (Reset, Reroute, Restore), which temporarily resets the agent's context trajectory to escape the absorbing region and then restores the historical context to effectively complete the task. Our approach yields 17-55% lower measured inertia across models relative to unmodified agents. These results suggest that changing the context can interrupt recurrence more effectively than directly steering the resulting activations. Our code is available at https://anonymous.4open.science/r/vlm-agent-defense-D076

Dynamic simulations are an entrenched way of gaining insight into the evolution of system dynamics. Their computational cost however is often prohibitively high, especially in cases of stochastic frameworks. Machine learning algorithms are especially suited as simulation surrogates. Nevertheless, they face some very distinct limitations. Firstly, the sheer dimensionality of these systems, however, precludes the use of traditional time series models who struggle with high dimensional feature spaces. Additionally, traditional time series focus exclusively on either long or short range effects, causing local or global drift given enough time. In this paper, we propose a framework that addresses those limitations. Our framework combines a Variational Autoencoder, with a convolutional or graph basis that reduces the dimensionality of the system. This latent vector is propagated in time using a Temporal Fusion Transformer model, which includes both long range and short range effect encoding, as well as static covariate support. We test our framework on three distinct cases, to prove its robustness and in all three we have achieved practically identical to the simulation results at a fraction of the time. Further, our framework is flexible enough to be adapted to any new system and provides an inbuilt uncertainty quantification for targeted experiment design.

Pretrained text-to-image models contain broad visual knowledge, yet they cannot reliably acquire or refine a specific visual identity from only a few references while preserving compositional control. Token-embedding methods are compact but often underfit identity, whereas adapter-based methods improve fidelity through persistent weight updates that can be costly to store and interfere when concepts are composed. We introduce V-Engram, a trigger-indexed external memory mechanism for Stable Diffusion 3.5. Each concept is assigned an explicit trigger that retrieves concept-specific memory, whose gated directions enter frozen text-encoder and MMDiT context states as relative residuals. Separating this memory from backbone adaptation enables prompt-selective and multi-concept access without merging model updates. Experiments show that V-Engram broadly matches DreamBooth-LoRA in overall subject fidelity while showing advantages in settings such as contextual subject preservation. Prompt-matched loading retrieves only matched entries, reducing most additional adaptation-state loading for a single-concept query. Qualitative results further demonstrate paired-trigger composition and same-class separation, while prompts without registered entries retain the frozen model's base behavior. Together, these results establish trigger-indexed memory as a modular interface for adding targeted visual evidence without rewriting the generator.

Large language models (LLMs) are increasingly deployed for natural-language logical reasoning, where the final answer is easy to check but the proof behind it is not. In natural-language logical reasoning, an intermediate conclusion should follow from its premises, and the resulting derivation should support the final answer. Existing methods lack machine-checkable verification of intermediate conclusions and answer-supporting proof dependencies, so they may assign credit to invalid or answer-irrelevant steps. We propose Proof-R1, an RL framework from formal verification that trains LLMs to construct verifiable proofs for natural-language logical reasoning. Proof-R1 admits a generated conclusion into the verified proof state only when the corresponding reasoning action satisfies the proof obligations through UNSAT-based machine-checkable formal verification. Proof-R1 also recovers the answer-supporting dependency closure to trace the proof structure of the final answer and align outcome credit with the proof dependencies. Experiments demonstrate that Proof-R1 improves answer accuracy across three logical reasoning benchmarks and four backbone models and outperforms training-free agents and training-based methods in terms of reasoning-process verifiability.

Exploring high-order correlations and long-range dependencies in brain networks holds significant value for both neuroscience research and clinical diagnosis. However, previous studies have lacked a unified integration of high-order and long-range dependency information in brain networks, and there is substantial redundancy behind various types of information. These issues limit their effectiveness in the diagnosis of brain diseases. To address this, we propose an Information Bottleneck-Guided Adaptive HyperGraph Transformer (IBAHGT). By incorporating the information bottleneck (IB) principle, this approach enables adaptive learning of high-order correlations and both short- and long-range dependencies within a unified framework for brain network analysis, achieving high-precision brain disease diagnosis. IBAHGT consists of three key components: an information bottleneck-guided adaptive hypergraph convolution, which introduces a novel hypergraph information bottleneck (HIB) principle to adaptively learn hypergraph message-passing weights between nodes and hyperedges, optimizes information flow and captures high-order information in brain networks that is maximally informative and minimally redundant (MIMR). The Transformer encoder captures global information within brain networks through the attention mechanism, specifically modeling short- and long-range dependencies. An information bottleneck-guided node-level adaptive fusion employs the IB principle to learn independent weights for each node, facilitating the fine-grained integration of high-order information and global information to obtain an efficient representation for downstream tasks. Extensive experiments demonstrate that the proposed method outperforms current state-of-the-art methods and can identify biomarkers for clinical applications.

Large language models (LLMs) are increasingly deployed to solve complex scientific and practical problems via iterative optimization. However, dynamically coordinating diverse search mechanisms as candidate quality, failure modes, and resource budgets evolve remains a critical open challenge. Targeted empirical diagnostics reveal that mechanism effectiveness is highly state-dependent. Motivated by this, we analyze how individual decisions drive final outcomes, decomposing the expected terminal improvement under a shared budget into cumulative decision opportunities minus cumulative selection losses. Guided by this opportunity-loss theoretical foundation, we propose OptiCom, a unified framework that represents LLM-driven optimizers within a shared configuration space: C=(A,Q,O,E,M,S), corresponding to artifact, query, operator, evaluation, memory, and strategy. Operating within this space, a fast LLM-based Optimization Controller dynamically composes immediate mechanisms through structured Action Packages, while a slower Strategy Adapter refines long-term selection preferences, operator weights, and templates based on accumulated trajectory feedback. Comprehensive evaluations across 32 benchmark groups demonstrate the superiority of framework: OptiCom achieves an average Max-score rank of 1.72 among 14 evaluated configurations, securing the top score in 23 groups. Ultimately, these results highlight the broad applicability and high extensibility of OptiCom as a general-purpose paradigm for robust LLM test-time scaling.

An agent that uses tools typically responds to what the user explicitly asks, yet completing the task may require information the user never requested. Work on proactive agents mainly studies whether and when an agent should act on its own, not what information it should pursue. We study a distinct axis of proactivity: its content. Horizontal proactivity pursues unstated information that the current context already identifies, and vertical proactivity pursues needs that only earlier evidence reveals. A need graph, recovered from a benchmark's own decomposition, records which needs depend on which, so both forms, and whether the agent stops at the right time, can be scored from a transcript without a model judge. To learn this behavior, we propose Q&D (questioner and drafter), which trains a questioner to prefer the question whose continuation retrieves more of the required evidence, with no reward model or judge. On held-out splits of three multi-hop question-answering benchmarks, at equal retrieval spend, the trained questioner improves both forms of proactivity over the same model, prompted, and outperforms a prompted model $15\times$ larger in the same role on two of the three, and the gain persists after controlling for question volume and length. Without further training, we place the questioner in an interactive customer-service agent with a simulated customer, where it completes more tasks while asking fewer questions, and in retail it outperforms the $15\times$ larger model with fewer follow-up turns from the customer. These results show that proactivity depends not only on whether an agent acts without being asked, but also on what it chooses to pursue and when it stops.

Proactive LLM agents can turn idle compute into useful support before users ask. Yet even correct work can misread user context, impose review costs, or undermine trust. This work proposes foundations for designing, realizing, and evaluating proactive LLM agents around three joint principles (3T): Task Capability, anticipating relevant needs and correctly performing useful work; Temporal Allocation, allocating compute according to resource availability and when results are needed; and Trust, sustaining users' confidence and appropriate reliance on the agent. We connect these objectives to a design space organized around five dimensions: task scope, anticipation horizon, activation trigger, processing timing, and intervention depth, and specify the situation and system modeling needed to support its choices, including user and environment representations, backbone LLMs, and agent harnesses. Lastly, we propose PROACTIVITY-GYM, a simulation-based evaluation testbed including multi-day scenarios, stateful environments, and persona-conditioned simulated users that can evaluate the consequences of proactive assistance across interactions. Evaluations across 23 model-harness configurations uncover substantial performance gaps across 3T and reveal that LLM judges often conflate task capability and trust. A human study with 30 participants demonstrates the importance of the joint 3T optimization: participants show sharp trust declines after intervention misalignment despite correct outcomes, and prefer sleep-time assistance, even when imperfect, to preserve ongoing focus. Together, these findings support designing and evaluating proactive agents through the joint consideration of useful work, compute allocation, and evolving user trust.

Non-injective mappings in neural networks map distinct inputs to the same representation, thereby implicitly inducing equivalence relations in the input space. However, the input differences eliminated by these mappings may still be required by downstream tasks, creating a mismatch between operator-induced indistinguishability and task-required distinctions. For non-injective linear operators realized in the current forward pass, their null spaces exactly characterize these invisible input variations. We propose Task-Relevant Null-Space Residuals (NSR), a general residual framework for non-injective linear mappings. NSR combines null-space component extraction from pre-mapping representations, member-level encoding and gating, and application-specific integration to exploit potentially task-relevant information under downstream supervision while preserving the original aggregation or merging rules. We evaluate NSR in two structurally different settings: token merging and graph aggregation. In token merging, NSR achieves higher semantic segmentation performance than the corresponding compressed baselines in 34 out of 36 evaluated configurations, with a maximum observed gain of 31.51 mIoU points under strong compression. In graph aggregation, NSR achieves 100% training accuracy on Tree-NeighborsMatch at depths d=2--6 across three backbones, alongside gains on heterophilic node classification and molecular graph regression. Together, these results support null-space residuals as a practical complement to non-injective linear mappings, enabling downstream models to learn from input distinctions invisible in the original operator's output.

Neural solvers offer efficient surrogates for numerical simulation of partial differential equations (PDEs). For time-dependent problems, strong one-step accuracy does not necessarily translate into reliable autoregressive rollout. We observe that a solver based only on physical-state modeling can achieve lower one-step error, whereas its spectral-only counterpart can become more accurate at later rollout steps. Motivated by this observation, we present Transolver-$\sigma$, a neural PDE solver based on joint spectral--physical subspace modeling. Within each block, adaptive physical-state interactions and spectral transformations are modeled in dedicated latent subspaces, whose responses are recomposed to enable information exchange between the two representations. Within the physical subspace, we introduce Slice-Residual Physics-Attention (SRPA), which preserves an explicit slice-space identity path while retaining learnable cross-slice interaction. In parallel, an axis-factorized Fourier operator captures global spectral structure. Across five well-established PDE benchmarks spanning steady-state prediction and time-dependent dynamics, Transolver-$\sigma$ achieves state-of-the-art with a benchmark-averaged relative error reduction of 33.4% over the strongest baseline for each metric, while consistently improving autoregressive rollout over single-operator counterparts. Transolver-$\sigma$ further delivers strong gains on coupled multiphysics systems and real-world fluid and combustion measurements from RealPDEBench, demonstrating its effectiveness beyond standard simulation benchmarks.

Laboratories often face a new molecular assay with 16-64 labels and a bank of predictors whose training data and parameters are unavailable. The practical question is which frozen outputs to include in a small local model. AssayRouter treats completed assays as pseudo-targets and labels each candidate by its post-fit utility: the reduction in held-out discovery loss when the candidate is added to the local target predictor. A shared regressor learns to predict this utility from candidate behavior on the support set, without source identity; on a new assay, one frozen ranking selects four sources and separate labels fit a convex combiner. We train only on completed ChEMBL-MT assays and evaluate 24 external regression assays across six frozen interface families. AssayRouter-C lowers strict four-call negative log-likelihood (NLL) by 0.0409 relative to Support-CV@4. Frozen candidate-label permutations confirm that candidate-utility correspondence carries the transferred information, and leave-one-interface-out training shows that the mapping generalizes to unseen predictor families. Completed assays therefore provide transferable supervision for scarce-label routing through frozen prediction interfaces.

Large reasoning models improve performance on challenging problems by allocating additional computation before answering, but longer reasoning does not always lead to better results and can introduce substantial redundant reasoning on simple problems. Conversely, aggressively shortening reasoning can degrade performance on difficult ones. Effective reasoning therefore requires dynamically deciding when additional computation is useful based on the reasoner's capabilities and evolving solution state. Existing approaches often rely on predefined budgets or intervention rules, retrain the target reasoner, or require additional supervision. We introduce MetaCtrl, a lightweight controller that adaptively regulates a frozen reasoner without predefined token budgets or reasoner retraining. We formulate reasoning regulation as a sequential metacognitive control problem: MetaCtrl observes the evolving reasoning trace and decides whether to continue, simplify, skip redundant steps, or conclude reasoning. It is trained directly with reinforcement learning using a reward that prioritizes correctness while favoring shorter trajectories among correct solutions, requiring neither supervised intervention trajectories nor problem-specific budgets. Across seven benchmarks spanning mathematics, science, and code, MetaCtrl consistently improves the accuracy of LRMs while reducing their reasoning length. On DeepSeek-R1-Distill-Qwen-7B, it improves average accuracy by 4.7 points while reducing generation length by 53.3%. Without further training, the same controller transfers to an unseen reasoner (e.g., Qwen3-14B), improving average accuracy by 2.9 points and reducing generation length by 50.3%. These results establish MetaCtrl as a plug-and-play controller for improving reasoning accuracy while substantially reducing inference-time generation. The code is available at https://github.com/binbin2xs/MetaCtrl.

Recent Recommendation Agents (RecAgents) offer a promising alternative by shifting recommendation to an active, user-side paradigm, where generative agents autonomously perceive external platforms, reason over user preferences, and execute decisions. However, existing RecAgents still suffer from two critical limitations: brittle item perception based on noisy and heterogeneous item pages, and inefficient long-context reasoning over extended user histories and multi-step interaction traces. To address these challenges, we propose a novel recommendation agent framework, termed as ReMem, that combines OCR-based multimodal perception with time-evolving dynamic memory. Instead of parsing raw HTML, ReMem observes item pages through screenshots and extracts structured multimodal information via an OCR tool, enabling a more humanoid and platform-agnostic perception mechanism. To support long-horizon preference modeling, ReMem further introduces a chunk-wise sequential memory update strategy, where the agent selectively maintains a fixed-size memory of informative historical interactions while processing arbitrarily long contexts with linear inference complexity and bounded context length. This design allows the agent to preserve evolving user preferences without relying on external memory modules or disrupting the standard autoregressive generation process. To enhance the dynamic memory instruction, we further develop a multi-memory GRPO variant, which propagates the final-answer advantage to all intermediate conversations that contribute to the final response. Extensive experiments on three datasets demonstrate that ReMem consistently outperforms state-of-the-art baselines, achieving an average improvement of 5.16\% across three recommendation agent tasks, namely searching, ranking, and judging.

Rubric-based evaluation is widely used to assess LLM-based systems by decomposing response quality into task-specific scoring criteria. However, automatically generating rubrics that reliably capture task-specific quality requirements remains challenging. We introduce Mubric, a mutation testing-guided approach to rubric generation. Mutation testing, a classic software testing methodology, evaluates a test suite by injecting faults into programs and checking whether the tests detect them. We draw an analogy between test suites and rubrics: if a rubric captures an important quality requirement, introducing a corresponding defect into an otherwise high-quality response should reduce its score. Mubric first mines common defects from real pairs of preferred and dispreferred responses and abstracts these defects into reusable mutation operators, each specifying how to introduce a particular type of response defect. For a new task, it applies relevant operators to a reference response, checks whether the injected defects reduce response quality, and uses insufficiently penalized defects to refine the rubric. We evaluate Mubric on 703 tasks across four representative domains against six advanced rubric generation methods. Mubric achieves the highest overall evaluation accuracy, outperforming the strongest baseline by 7.48 percentage points.

Conflicting emotional cues can be individually valid: a subdued voice may reflect a blocked goal while a smile satisfies a social obligation. Their interpretation depends on what the event means to the person. We introduce VISTA (Value-Informed Semantic Trust Arbitration), a learned seven-field appraisal interface that conditions modality arbitration on concerns, event relations, and expression conditions while retaining a joint-evidence residual. A log-odds decomposition separates emotion expectation from cue diagnosticity, motivating an interface that lets appraisal change how evidence is interpreted. With a shared Qwen2.5-Omni-7B backbone and matched training examples and steps, VISTA reaches 64.5% conflict accuracy on CA-MER, improving on modality gating by 2.5 percentage points on conflict and 0.2 on consistency. Shuffling appraisal across scenes or removing its decision connection reduces this benefit. A common frozen-backbone probe reaches 0.600 macro CCC for appraisal readout, compared with 0.505 for emotion-only fine-tuning. Evaluations across five benchmarks connect recognition under increasing conflict with appraisal readout and downstream decision use. Together, the analyses and experiments support scene-specific appraisal as an intermediate representation that helps interpret conflicting emotional evidence.

On-policy distillation, where a student learns from a stronger teacher's feedback on its own outputs, is a common way to pass reasoning to smaller models. We analyze what it transfers at small scale, distilling Qwen3-8B into Qwen3 4B, 1.7B and 0.6B students, in thinking mode (reason at length, then end the reasoning and answer) and, for comparison, in non-thinking mode (no separate reasoning phase). Long reasoning needs two abilities, solving a problem and knowing when it is solved, and we find that distillation transfers the first, but in thinking mode not the second. Solving improves at every size, up to two ceilings, which we measure comprehensively across both modes and all student sizes: a student's single attempt never exceeds what it could already reach in many attempts before training, and the smaller the student, the further it stays below the teacher. Stopping is where the modes part. In non-thinking mode every student keeps stopping; in thinking mode students stop ending their reasoning early in training, and the smaller the student, the less of this ability survives: the teacher signals a stop almost only where a student already ends its reasoning, so distillation teaches no new stops; it only keeps the student's existing stops that land on a right answer, and a weak student has few such stops. The smallest students often reach the right value but do not commit to it: they either rarely mark it or mark it and write past it. Together, these results describe how small students behave under on-policy distillation, and a diagnostic that separates answer marking, correctness and stopping.

Vision-Language Navigation (VLN) requires agents to continuously ground task progress from long-horizon instructions and partial egocentric observations. Existing VLM-based navigation agents typically reason only over available observations and may remain confident even when task-relevant evidence is missing. For example, an agent may confidently proceed forward and get lost even though the landmark indicating the next turn lies outside its current field of view. We term this failure mode Progress Myopia: the agent fails to recognize unreliable progress grounding and continues acting on insufficient evidence. To address it, we propose SeekVLN, an evidence-seeking framework that couples semantic progress reasoning with active acquisition of task-relevant observations. SeekVLN is trained in two stages: First, Future-guided Reverse Generation (FRG) uses future expert actions to augment offline expert trajectories with supplementary views and evidence annotations. Supervised fine-tuning on these trajectories establishes a prior for evidence seeking and progress reasoning without additional expert interaction. However, imitation alone does not reveal whether seeking improves subsequent navigation. We therefore introduce Counterfactual Contrastive Policy Optimization (C2PO) for reinforcement fine-tuning. By comparing each evidence-seeking branch with a counterfactual direct-navigation branch from the same state, C2PO uses a contrastive reward to assign credit to seeking decisions based on subsequent navigation benefit. Experiments on simulated benchmarks show that SeekVLN achieves state-of-the-art performance, improving success rate by 12.7% and 7.5% over the base model on R2R-CE and RxR-CE, respectively. Both simulated and real-world evaluations exhibit human-like evidence-seeking behaviors for more reliable progress grounding.

Generating optimization instances that are both feasible and computationally challenging is crucial for benchmarking solvers and training learning-based optimization algorithms. Existing non-LLM generators rely on seed instances or parameter tuning, resulting in high test-time computational cost, while existing LLM generators lack explicit hardness measures. Recent reinforcement learning methods with verifier feedback evaluate only binary correctness, which is misaligned with generating challenging problems. We note that an optimization solver reports the cost of solving at several stages of its pipeline, and leverage this to design a reward that scores both the solvability and the hardness of generated problems, measured by branch-and-bound nodes and post-cut relaxation gaps. Our key idea is a challenger-solver asymmetric self-play approach, where an LLM challenger generates progressively harder instances and the solver verifies feasibility and hardness, so no seed or training MILP instances are required. We fine-tune Gemma-4-12B and Qwen3.5-4B with GRPO and a size curriculum into OptiScribe-12B and OptiScribe-4B, which generate feasible yet challenging MILP problems from natural language instructions. On capacitated facility location and max-cut, OptiScribe-12B raises median SCIP search nodes by 1.7-5x and post-cut gaps by 1.1-1.7x over its base model and improves the feasibility rate on facility location by 9-19 points, while OptiScribe-4B raises median nodes by up to 15.6x. The problems cover a wider difficulty range than public benchmarks of the same size, follow instructions on density and difficulty, and can tune solver settings for families that public libraries lack. These results indicate that optimization-specific rewards, used in self-play mode, can teach LLMs to generate high-difficulty optimization benchmarks. We will release our code and models publicly on acceptance.

Large language model (LLM) routing aims to assign each query to the most suitable model from a heterogeneous candidate pool, improving the quality--efficiency trade-off of LLM inference. Existing routers are typically learned through local fitting: a router is optimized for a particular query workload and candidate pool, and often requires additional supervision or retraining as the routing environment changes. We ask whether LLM routing can instead be approached from a foundation-model perspective, learning a reusable routing capability that generalizes across tasks, candidate models, and deployment conditions. To this end, we introduce RouteFM, which learns to characterize anonymous candidate models from behavioral context and infer their target-specific capabilities, rather than binding routing decisions to fixed model identities or a single environment. Through episodic pretraining across heterogeneous routing environments, this capability can be reused by a frozen router and adapted to new environments through context alone. Experiments demonstrate transfer across changes in domains, modalities, candidate pools, and context budgets, with the largest gains when behavioral evidence is limited. On MMR-Bench, which is excluded from pretraining, RouteFM outperforms the strongest baseline by 2.23 quality points with only eight observations per candidate. These results support moving LLM routing from repeated local fitting toward a pretrain once, route anywhere paradigm. Our code is publicly available at https://github.com/LAMDA-Model-Reuse/RouteFM.

On-policy distillation (OPD) trains students using teacher feedback on their own sampled responses, yet how prompt choice shapes transfer across teacher-student pairs remains poorly understood. We systematically study prompt quantity, source, and selection across RL- and SFT-continuation pairs and cross-model settings. We find that OPD can be highly prompt-efficient: a few prompts can approach large-pool performance, with four DAPO prompts matching the observed mathematics score of 3,840 DeepMath prompts. However, prompt utility is relational rather than intrinsic: changing only the teacher can reverse the relative effectiveness of mathematics and code prompts. To characterize these transfer differences, we analyze parameter and functional changes across prompt supports and model pairs. Functional alignment with the teacher varies across supports and target tasks; in continuation pairs, teacher-aligned prediction changes can coexist with weak parameter alignment. Continued OPD on effective supports can restore performance after unfavorable transfer. Finally, targeted selection does not consistently outperform uniform random sampling, and filtering out a source that performs poorly alone yields no consistent gain across three paired support draws. Overall, our results distinguish prompt efficiency from prompt interchangeability and show that effective data choice depends on the teacher-student pair and target capability, with random sampling providing a competitive baseline in the studied settings.

World-Action Models (WAMs) couple action generation with predictions of how physical interactions unfold. However, current post-deployment learning paradigms typically improve behavior without requiring better world predictions. Especially in dexterous manipulation, small execution errors can compound in high-dimensional action spaces, hindering policy improvement and pushing interactions beyond the world model's training distribution. Motivated by this, we propose Direct Experience World-Model Optimization (DEWO), a post-deployment learning paradigm for WAMs that, alongside action imitation, refines world representations through visual experience to better condition action generation. Specifically, it identifies interaction turning points and learns from successful and failed futures to support classifier-free guidance. An additional value head estimates task progress from video representations and activates guidance when progress stalls during inference. Across five DexJoCo tasks, DEWO improves average success across all three WAM formulations. Ablations show that visual supervision from successful and failed continuations improves both prediction and control beyond action supervision alone. On four real-world tasks across Wuji and Sharpa, 3 x 3 grid evaluations show that two rounds of deployment learning increase success from 51.0% to 71.7% in cells with at least one initial success, a gain of 20.7 percentage points. These findings support continued predictive learning for improving control through deployment experience, making world modeling an active part of WAM adaptation.

Large language model (LLM) routing reduces serving cost by assigning each query to an appropriate model while preserving response quality. Learning such a router, however, often requires executing multiple candidate models on historical queries to collect query--model quality feedback, creating a nontrivial supervision cost before deployment. Existing work largely focuses on serving-time efficiency, overlooking whether the resulting savings are sufficient to recover this upfront expenditure. We further observe that routing quality often saturates well before all query--model feedback is collected, suggesting that dense supervision can be economically over-provisioned. We propose SaveRouter, a sparse-supervision routing framework that selectively acquires informative model feedback and shares capability information across related queries, while retaining query-level refinement for fine-grained routing. We evaluate routing by jointly accounting for supervision expenditure and subsequent serving-time savings. Across four routing benchmarks, the main setting uses only about 33--41% of available training feedback while maintaining competitive or better routing quality, and reduces the break-even deployment volume by approximately 1.9--9.5 times compared with the fastest conventional router. Further analysis shows that acquiring more supervision is not always economically preferable: the supervision level that minimizes serving cost can differ from the one that achieves the earliest payback. Our code is publicly available at https://github.com/LAMDA-Model-Reuse/SaveRouter.

Supervised causal discovery learns to infer causal structure for a new dataset from training datasets paired with structural labels. These training pairs are typically simulated, making the simulator both a source of supervision and a carrier of assumptions about causal graphs, mechanisms, and noise. Understanding the resulting predictions therefore requires examining how these assumptions supplement the information available in observational data, which may be compatible with multiple causal graphs. This paper examines that relationship across representative methods available through June 2026. We organize these methods by prediction target, prediction granularity, encoder, structural decoder, and training regime to relate what each method predicts to how it uses data and simulator-based supervision. Using this framework, we distinguish two questions: whether the target is identifiable under the assumed model class, and whether a trained predictor generalizes beyond its training distribution. Restrictions on mechanisms and noise can make otherwise ambiguous causal directions identifiable, but predictive accuracy under those restrictions does not establish transfer when they change. This distinction motivates evaluation that matches metrics to the identifiable graph target and tests changes in graphs, mechanisms, and noise between training and deployment. Extending such evaluation to real data also requires documenting the external causal evidence and uncertainty behind benchmark reference graphs. Together, these analyses guide method comparison and identify open questions in transfer, test-time adaptation, and uncertainty assessment.

When a user changes their mind partway through a task, an agent that has already split the task into sub-goals and paid for tool calls must decide, per cached sub-result, whether to keep, patch, or discard it (salvage), restarting wastes valid work and continuing unchanged answers the old question. Our main finding is that salvage quality is a matter of role design rather than model capability: a language model asked the keep/patch/discard question one node at a time is unreliable, but asked to classify the revision once, with a deterministic layer propagating the decision, it reaches the cost-optimal oracle on all three models tested, from two vendors. Modeling the plan as a commitment hierarchy and the intent change as a belief-revision operator with AGM style postulates, we prove that no policy observing only a node's local view can be both safe and cost optimal, while the single classification design is both. Across three environments the policy recovers the full achievable savings, 43% cheaper than restart, at 100% correctness.

Underwriters in commercial Property and Casualty (P&C) insurance spend 30 to 40% of their time on administrative work rather than risk judgment, and a single submission takes about 40 minutes by hand. We present Governing the Edge, a multi-agent framework for that layer, organized around a data-residency constraint: sensitive submission data must not leave the perimeter. Eleven agents and two deterministic control nodes, one a human-escalation interrupt, form a 13-node LangGraph workflow across two tiers. Agents touching raw submissions run locally on Gemma 2, each bound through a tool interface logging every invocation (synthetic stubs in the current prototype); only anonymized scores and non-identifying fields cross to Claude Sonnet in the cloud. Compliance rules are encoded as conditions on graph edges, so a non-compliant submission never reaches pricing. We walk through one full scenario end to end, a commercial auto submission whose principal driver carries serious violations, showing every agent call, every tool invocation, and the resulting routing decision. The framework processes a 40-minute manual submission in a few minutes on a single edge device, dominated by serialized on-device inference. On the hard-stop violation tier the framework enforces every rule correctly and reproducibly, since hard stops are deterministic predicates over fields extracted at temperature 0; overall compliance accuracy across the 20-scenario benchmark is 70%, with the remaining gap concentrated in softer, judgment-based tiers. It keeps a complete audit trail and respects the privacy boundary throughout. We release all code, compliance rules, tool stubs, and synthetic datasets.

Enterprise artificial-intelligence agents increasingly call tools, modify infrastructure, and process protected data, creating a need to separate action generation from action authorization. This article presents VeriWeave Govern, a deterministic runtime governance layer that evaluates structured agent actions against versioned policies, validates typed evidence, applies fixed deny > review > allow precedence, routes consequential actions to accountable human review, and records replayable tamper-evident audit state. GovernBench evaluates the design over 30 independent seeds and 60,000 oracle-labelled cases spanning five enterprise domains, adversarial evidence, out-of-distribution actions, and temporal policy evolution. VeriWeave achieves 0.9888 mean accuracy, 0.9836 macro-F1, zero observed aggregate false allows, and zero observed Governance Attack Success Rate on the evaluated cases. Six ablations show that evidence gating, deny precedence, out-of-distribution fail-safe behavior, human review, contradiction handling, and temporal replay contribute complementary safety. The deployed API additionally passes 12/12 end-to-end scenarios and a 40,040-request concurrency matrix with zero failures. A separate 150-case EU/Austria regulation-grounded evaluation uses frozen predictions and two independent blinded human annotators, who agree on all decisions. On this set, deterministic engines remain conservative, while a Gemma 4 31B comparator aligns more closely with the human consensus. The results expose a measurable safety--utility trade-off and motivate evidence-aware, replayable governance as an independent control plane for enterprise agent execution.

Industrial standards often encode the same engineering concept through incompatible hierarchies, identifiers, roles, and structural constraints, so the nearest lexical or embedding match can still be technically inadmissible. This article presents CRASM, a deterministic constraint- and role-aware semantic mapping method, and CRASM-Gate, its selectively model-assisted extension. The framework separates standard-specific canonicalization, bounded retrieval, deterministic rules, destination-versus-origin role interpretation, semantic and structural ranking, ambiguity refusal, and target validation. CRASM-Gate adds a confidence/disagreement gate that may invoke a candidate-constrained large language model, while final authority remains with deterministic validation. A controlled artifact covers six directed industrial-standard pairs, three difficulty levels, and ten configurations, yielding 14,400 sample-level decisions. With a fixed local model endpoint, CRASM-Gate reaches mean F1 0.9938 and an implemented structural-validity rate of 1.0000; deterministic CRASM reaches 0.9931 without generative-model calls; and the model-only baseline reaches 0.5347. Relative to the model-only baseline, CRASM reduces top-1 errors from 670 to 10 while exhibiting 0.0688 s/sample rather than 24.5263 s/sample observed latency. CRASM-Gate improves CRASM by one additional correct decision out of 1,440, with observed latency increasing to 15.7666 s/sample. The results support a deterministic-first interoperability architecture in which model assistance is optional, measurable, candidate-bounded, and unable to bypass structural validation.

Persistent memory creates a control problem that retrieval relevance alone does not solve: a memory can remain highly useful after an update, deletion, or revocation makes it inadmissible for the current answer. We formalize this as a separation between utility and authority. A fixed finite penalty applied to an unnormalized utility score cannot guarantee exclusion under arbitrary positive-affine reparameterization of that score; by contrast, rank-normalized compensation is scale-invariant and therefore forms a stronger empirical comparator. Our prospectively frozen TIDE/LongMemEval primary was quarantined before a valid HELDOUT comparison because the materialized TIDE adapter conflated historical age with query-relative inadmissibility and the aligned LongMemEval split left no DEV set for the predeclared penalty selection. We therefore report a post-primary replacement diagnostic on Memora Remembering, where update/delete operations provide item-level forgetting state. On Qwen3-8B, DEV selected lambda = 0.6 from a ten-point normalized SOFT family. Across 185 HELDOUT units in 28 dependency clusters, HARD exclusion yields 4.04% balanced construct error versus 19.66% for locked SOFT, a paired difference of 15.61 points with a 20,000-replicate cluster-bootstrap 95% interval of [13.07, 18.76]. The effect is driven primarily by forgotten-value leakage while current-value recall is preserved. This is same-Q operator-comparison evidence, not a universal claim that scalar control fails, not an evaluation of learned authority inference, and not an independent downstream-harm endpoint.

Tool-using language-model agents can decide that an action is permissible and execute it only after security-relevant state has changed. We study this proposal-to-commit gap and introduce BSC-R, a deterministic effect-boundary mechanism that binds a single-use commit authorization to the exact action and to a semantic projection of the authorization state that justified it. On 2,847 attacked AgentDojo episodes, the boundary kernel preserves the unprotected agent's behavior exactly (80.576% utility; 1.616% attack success). On 10,302 frozen proposals, it accepts every unchanged commit and rejects every instance of ten prospectively specified changed or replayed classes. In an independently generated boundary-drift experiment, full joint binding commits 0/4,403 invalid contexts while retaining 5,899/5,899 valid contexts. A prospective external evaluation on the 3,460-scenario CONTINUITY suite retains all 700 benign cases, prevents 1,200/1,200 represented non-replay invalid effects, handles 160/160 replay lifecycles correctly, and withholds 200/200 ambiguous no-release cases. The broader external suite also exposes the method's limit: across all 2,560 attacks, BSC-R has a 25% invalid-effect commit rate versus 0% for CONTINUITY. The result is therefore a scoped commit-time consistency mechanism, not a universal agent-safety claim.

Skills equip LLM agents with professional knowledge and guidance to complete long-horizon and complex tasks. Although skills have been widely adopted in recent agent paradigms and harnesses, how to synthesize reliable training data and how to train agents for skill use remain underexplored. In this work, we propose SkillGym, an automatic pipeline to build verifiable environments, collect trajectories, and train skill-use agents. SkillGym first crawls a large volume of skills from the internet, then keeps those whose workflows can run reproducibly offline. A builder-reviewer pipeline is used to construct difficulty-controlled tasks, spanning four task types, each with a reference solution and an executable verifier. With this pipeline, we build 6.8k environments and collect 19k verified successful trajectories for supervised finetuning. Finetuning on these trajectories improves LLMs of different families and sizes, from 2B to 122B parameters across four skill-use benchmarks; Our Qwen3.5-9B SFT model outperforms the 397B untrained model on two of them. Further analysis shows that training teaches agents to invoke skills, raising the rate of reading the relevant skill from 28% to 96%, and that the gains hold across reasoning structures, extending to task types that form a minority of the training data and to skills held out from training

Content-generation agents continuously receive impressions, clicks, conversions, and negative feedback from recommendation systems, providing real-world outcome signals for memory evolution. However, these signals are delayed and noisy, confounded by audience composition, placement, and recommendation policies, and may result from the combined influence of multiple memories, making accurate attribution difficult. Existing methods rely primarily on immediate feedback or semantic retrieval and therefore struggle to reliably translate recommendation outcomes into memory fitness. To address this challenge, we propose TIDE (Trajectory-Informed Directed Memory Evolution), an external memory evolution framework driven by delayed recommendation feedback. We further introduce Memory Evolution Gain (MEG), which measures the utility improvement of evolved memory over a no memory baseline on strictly future tasks. TIDE treats memory as a capacity-constrained population of experiences: temporal and semantic credit assignment estimates contextual fitness, while responsibility credit distributes outcome signals according to the memories referenced during generation. These signals are then used to reinforce, crossover, mutate, or evict memories. On an e-commerce membership marketing content-generation agent, TIDE achieves a +7.75-percentage-point MEG in offline temporal replay and significantly improves both unique click-through rate (UCTR) and activation rate in an online A/B test. On a delayed-label benchmark, TIDE achieves the lowest mean absolute error (MAE) and root mean squared error (RMSE) and the highest MEG among the compared methods, demonstrating its effectiveness.

Users of language-based assistive agents often make ambiguous requests. In response, an assistant can either directly act on its interpretation of the request --- risking misalignment with the user --- or ask a clarifying question. Which option is the most safe and helpful? A common approach is to ask questions that minimize uncertainty about the user's intent until a threshold is reached. However, this neglects the impact of uncertainty reduction on downstream performance, the costs of asking versus acting immediately, and the possibility that users may provide corrections without being asked. To navigate these trade-offs, we introduce Rational Enquiry via Value-of-Information Reasoning (REVOIR). REVOIR makes clarification decisions via inference-time reasoning about the value-of-information of a question, which captures the expected improvement in task reward due to the answer received. In two assistive tasks --- ambiguous question answering (CondAmbigQA) and preference-aligned household task planning (ADAPT) --- we show that REVOIR achieves greater success with fewer questions than approaches based on prompting, chain-of-thought, fine-tuning, or information gain, improving preference satisfaction on ADAPT by 13-15% over a fine-tuned clarification policy while requiring no training and asking five times fewer questions. Furthermore, when the assistant can receive cheap user corrections after acting, REVOIR naturally infers that asking questions is not always efficient, demonstrating the adaptivity of our approach. In contrast, we find that vanilla reasoning agents fail to adaptively clarify user requests, and request fewer clarifications as reasoning effort increases.

LLM agents accumulate interaction histories that grow linearly with task length, causing quadratic inference cost scaling and performance degradation from attention dilution. Existing context-compression methods learn what to discard offline: by contrastively optimizing guidelines, distilling compressors, or training compression policies. This incurs a substantial cost. Further, the compression policy is learned a priori and is not dynamically conditioned on the evolving test-time trajectories. In this paper we ask a complementary question: Which past interactions causally shape the agent's future decisions? We recast context compression as a causal decision preservation problem over discrete interaction units and introduce FOCUS, a training-free context compression framework that operates entirely at test time. Our method requires no offline data collection or fine-tuning, and is architecture-agnostic, attaching to any closed-API frontier model as a modular compression layer. We evaluate FOCUS on diverse agentic benchmarks including API and tool-calling, QA, web domain and multi-turn dialogue. Our method establishes new state of the art performance, cutting peak context by up to 48% and dependency by 73% while improving task success by up to 8.9 percentage points over uncompressed execution.

Uncertainty estimates tell us how unsure a model is, but not why. Without knowing which parts of an input influences a model's uncertainty, we cannot tell whether that uncertainty score depends on input features that are relevant for the task. We study this problem in randomset classifiers built using pretrained language models. These classifiers assign probability to individual answers and to groups of answers, producing lower and upper probabilities for each answer; The difference between these probabilities, called credal width, is used to represent epistemic uncertainty about an answer arising from limited training data. We propose XU-RS, a framework that attributes an answer's credal width to the input tokens (words or word pieces) supplied to a language model. XU-RS uses Expected Gradients (a standard feature attribution method) to estimate how input tokens contribute to credal width. The proposed framework is evaluated on a MedQA dataset using SmolLM3-3B and Llama-2-7B models, demonstrating that setting the embedding of a token ranked highly by XU-RS to zero (zero-masking) causes larger changes in credal width than zero-masking randomly selected tokens. In addition, we show that normalisation can cause other answer groups to influence an answer's width, reveal how token attribution can mask numerical errors, and provide diagnostic checks to verify whether a token ranked highly by XU-RS meaningfully explains model uncertainty.

Self-supervised pretraining reshaped prediction in language and vision, and brain foundation models (BFMs) inherited its promise. Representations learned from large unlabelled corpora should capture individual functional dynamics and generalise across cohorts. However, kernel ridge regression (KRR) fitted on functional connectivity (FC) matrices still predicts individual phenotypes more accurately than any BFM we tested. In this paper, we show that KRR is weighted by the eigenvalues of the FC which are miscalibrated for phenotype prediction. We apply an efficient spectral filter to recalibrate the eigenvalues of each subject's FC matrix, enabling the model to exploit more inter-individual variance. Across the 5 datasets, 11 parcellations and 6 prediction targets we tested, we match or exceed the KRR baseline. Based on this finding, we then pretrain a small encoder model on about 4,000 hours of fMRI from 162 open datasets, whereby we align the pairwise similarities between the embeddings of recording snippets with those between the recalibrated connectomes. Our model performs on par with the best of the 6 published BFMs we tested while having an order of magnitude fewer parameters. Our encoder performs better than FC on short scans and in smaller cohorts, especially in fingerprinting. We release the pretrained model weights, the code and the pretraining data, preprocessed and parcellated.

Latent world models often struggle with long-horizon planning despite accurate short-term predictions. Recursive rollouts accumulate errors, while distance concentration in high-dimensional latent spaces can weaken goal discrimination. We introduce the Dual-Latent World Model (Dual-WM), which separates local execution and long-range planning through distinct state representations and dynamics models. The low-level model predicts action-conditioned transitions, while the high-level model uses learned macro-actions to plan over longer temporal spans. We also propose Long-Horizon Representation Learning with Weighted Rollout (LoRe), which supervises self-generated predictions at both levels. An analysis of recursive error propagation motivates exponential horizon weights with separate decay rates for the two temporal scales. During planning, the high-level model generates latent subgoals that the low-level model refines into actions for precise execution. We evaluate from-scratch Dual-WM on five goal-conditioned visual control tasks against the task-wise strongest baselines without actor-guided proposals. At goal offsets of 50 and 100 environment steps, mean success increases from 75.9% to 84.4% and from 61.4% to 69.5%, respectively. At offset 100, Dual-WM outperforms these baselines on all five tasks and improves mean success over LeWM by 30.8 percentage points. Ablations and supporting analyses provide evidence of more informative representations for goal evaluation and greater consistency under recursive prediction. These results highlight the value of separating temporal roles and training across multiple horizons for reliable latent planning. Our core implementation is available at https://github.com/DeLin1001/Dual-WM-Official.

LLM agents are increasingly expected to support enterprise workflows, where tasks often involve missing information, uncertainty, feedback, and long-term trade-offs. However, existing enterprise and financial benchmarks mainly test static capabilities such as information extraction, numerical calculation, domain knowledge, and financial QA, leaving interactive and long-horizon decision-making underexplored. To bridge this gap, we introduce EnterpriseBench, a benchmark that evaluates LLM agents across this spectrum, from static question answering to dynamic decision-making. Specifically, EnterpriseBench reorganizes existing enterprise and financial QA datasets into a unified foundational suite annotated by capability and difficulty, and introduces three professional interactive settings: Consulting, based on management-consulting-style business cases for client problem diagnosis through multi-turn information seeking; the Beer Game, adapted from a classic supply-chain management simulation for inventory control under delayed feedback; and Enterprise Digital Twin, a project-based business simulator for workforce, risk, and project planning. Experiments with nine agent methods under four backbone models show that current agents have not yet achieved stable, comprehensive, and cross-task reliability in enterprise scenarios. These results show that EnterpriseBench provides a practical benchmark for evaluating LLM agents in realistic enterprise strategic reasoning and decision-making.

Experienced professionals know more than just facts and conclusions. They know which cues matter, why a judgment is reasonable, and which action to take. Routine work records often leave out this tacit knowledge, making it difficult for Large Language Model (LLM) agents to use professional experience effectively. We introduce KUPAS MASTER, an experience engineering platform built around nine-layer cognitive corpus construction. It turns heterogeneous work records and practitioner interviews into traceable, reusable experience corpora for agents. Six case elements preserve the task process: context, cues, judgment, action, boundaries, and outcomes. Nine-layer cognitive corpus construction organizes tacit experience along nine extraction dimensions and stores the resulting assets in six libraries: rules, constraints, best practices, negative examples, corner cases, and skills. Semantic alignment, individual experience distillation, organizational consolidation, and cross-review preserve source evidence, conditions of use, and unresolved disagreements. The platform packages these assets into callable skills with explicit inputs, steps, dependencies, and stopping conditions, connecting experience collection to task execution and evaluation feedback. Using authorized samples from 20 randomly selected practitioners, the platform processed 1,576 source files into 23,024 individual experience records and 13,113 organizational assets. The evaluation spans multiple professional domains. Under common task inputs and scoring criteria, the base model, raw corpus retrieval-augmented generation (RAG), and KUPAS MASTER agent scored 70.63, 79.75, and 89.58, respectively. The KUPAS MASTER agent improved on raw-corpus RAG in all seven scoring dimensions. The platform provides a practical path from individual tacit experience to organizational knowledge and agent capabilities.

Adaptive Cruise Control (ACC) systems are typically calibrated for an average driver, often resulting in a mismatch between vehicle behavior and individual expectations during time-critical maneuvers such as highway overtaking. When the ACC is perceived as too conservative and inconsistent, drivers intervene through throttle overrides, providing implicit feedback on the system's behavior. This paper reframes these override actions as human-in-theloop supervisory signals and proposes a data-driven framework for personalized vehicle adaptation, termed Context-driven Personalized ACC (CoP-ACC). Rather than relying solely on end-to-end regression, which tends to over-smooth dynamic responses, we introduce a hybrid pipeline combining: (i) unsupervised hierarchical clustering to extract representative acceleration profiles from override events; (ii) a context classifier that maps pre-maneuver driving conditions to the appropriate profile; and (iii) a residual regressor that refines the selected profile into a smooth, personalized acceleration profile tailored to the immediate context. Evaluated on real-world public-road data against a withheld forced-ACC baseline, the approach demonstrates high reconstruction fidelity and generates acceleration profiles that tend toward the driver's expected behavior in potential override contexts. The results highlight the potential of learning from shared-control overrides to enable anticipatory, personalized ACC behavior, reducing manual interventions and improving ride comfort.

Autonomous agents have made rapid progress in general-purpose computer use, but reliable automation of professional industrial engineering remains out of reach, as engineering workflows demand reasoning over geometric and physical constraints and dependencies preserved across software and design stages. We present EngiWorld, the first benchmark structured around the complete design loop: 1,301 expert-curated tasks spanning 6 engineering domains (CAD, CAE, CAM, BIM, EDA, and 3D visualization) and 26 professional software platforms, with both GUI and CLI interfaces and 6 task types ranging from software-selection to open-ended tasks. We further introduce an artifact-centric evaluation methodology built on a unified domain-verifier suite, which programmatically checks the geometric validity, physical feasibility, and rule compliance of final and intermediate artifacts, and scores quantitative design tasks continuously by specification attainment rather than binary success. Evaluation of seven frontier models reveals a substantial capability gap: the strongest model achieves an EngiScore of only 44.3, and just 3.6% of multi-software attempts succeed. EngiWorld provides the first rigorous foundation for measuring progress toward agents that operate professional engineering software end to end.

Active test-time adaptation (ATTA) improves robustness under distribution shift by updating a deployed model during inference while selectively querying supervision. However, most existing ATTA methods implicitly assume that supervision can be requested for every incoming test batch, which can incur substantial annotation cost over long test streams. In this work, we introduce \emph{budgeted ATTA} in which labels are available for only a fraction of test batches. This formulation shifts the central challenge from deciding \emph{what} to label within a batch to deciding \emph{when} supervision should be applied over time. To address this challenge, we propose a budget-aware approach \emph{WISE-ATTA} that allocates supervision over the test stream based on lightweight signals computed online, prioritizing periods where supervision is likely to be most useful. When a batch is selected for supervision, we further employ a drift-based sample selection criterion that targets samples exhibiting ongoing, unconverged adaptation dynamics, enabling effective updates from a single labeled example. We evaluate this approach on synthetic corruptions (ImageNet-C) and natural distribution shifts (ImageNet-R/K/A). Across settings, WISE-ATTA achieves competitive or improved performance compared to recent ATTA methods while requiring substantially fewer labels. Overall, we find that the timing of supervision is a key, yet underexplored, aspect of active test-time adaptation. Code: https://github.com/Muhammad-Huzaifaa/WISE-ATTA

Language models expose internal signals that predict whether an answer is correct, readable from a single forward pass of a frozen model without additional generations. Yet existing probes often commit to one signal family or layer and can be brittle under distribution shift; in retrieval-augmented settings, many specialized detectors instead target passage faithfulness, which can diverge from correctness when retrieved evidence is unhelpful or conflicting. We therefore ask where answer correctness is readable, which internal signal families carry it, and how they should be combined. We search over hidden states, token probabilities, residual-stream features, attention, and their fusion, treating the selected readouts as a predictive measurement rather than a mechanistic localization. We run this analysis separately in closed-book and with-context settings, since context can change which readouts are informative. A consistent anatomy emerges: correctness concentrates in the answer span, recovered from the answer tokens even under retrieval, and the families carry it complementarily, so fusing them helps most out of distribution, where a single signal is weakest. The protocol is effective across two backbones and gates a retrieval controller as one downstream use.

Human social behaviour is not a collection of independent motions, but a jointly organised process in which group dynamics and individual variation continuously shape one another. Yet existing social motion models often prioritise plausible trajectories while leaving interaction state implicit, limiting their ability to transfer across groups, tasks, and partial-observation regimes. To address this gap, we introduce Bilevel Representations for Agent Interaction Dynamics (BRAID), a hierarchical sequential latent-variable model for generative multi-person interaction. BRAID explicitly formulates social motion generation as a meta-transfer learning problem: shared interaction priors are learned across datasets and adapted through arbitrary context sets of observed people and joints. The model represents each scene through a group-level latent state that captures shared interaction dynamics and person-level latent states that capture individual behaviour conditioned on the evolving group context. This modelling choice enables coherent generation under full, sparse, or partial observations while exposing compact social-state vectors that can serve as an interface for downstream embodied-agent systems. We evaluate BRAID under a unified SMPL-based representation on social forecasting, tracking and in-filling, and response generation, using metrics that assess not only reconstruction accuracy but also realism, diversity, temporal alignment, and interpersonal coordination. We further analyse the hierarchical latent space, showing that it captures separable group- and individual-level structure.

We introduce Context Language Models (CLMs), language models that natively manage their own context. We implement this by treating the context as a file and allowing the model to make unrestricted updates to this file. This allows the model to learn what is most important to maintain in context, and naturally extends to multi-agent systems where multiple agent contexts coexist as files. Building CLMs zero-shot with existing models outperforms SOTA context management strategies across a variety of tasks: 11.4% higher accuracy with 21.5% fewer FLOPs on BrowseComp-Plus, 5% higher scores with 59% fewer FLOPs on 12-hour EdgeBench, and 65% greater improvement with the same compute on a 24-hour multi-repository agent-swarm task. Moreover, by shifting context management from external harness control to intrinsic model behavior, CLMs naturally enable both in-context and parametric learning of context-management strategies. We show that CLMs can be steered with natural-language instructions evolved through a standard skill-optimization loop, improving held-out accuracy by up to 35.9 points on a context-management task while reducing compute. We also introduce an online reinforcement learning method for CLMs, improving Qwen3.5-9B performance on BrowseComp-Plus by 47.6% while using 12% fewer FLOPs. Finally, we co-design Suffix Cache Reuse for CLM serving, further reducing server-side compute by 35% relative to standard SGLang at matched performance.

Seizure detection and prediction from EEG are clinically important but challenging because seizures are rare, temporally localized, and propagate as coordinated events across multiple channels. Recent dynamic graph neural networks model this by running a temporal model over a sequence of per-time-step pairwise channel edges. However, this pairwise construction misses the spatiotemporal coupling that constitutes a seizure, at substantial training cost. We propose HyBrain, which summarizes spatiotemporal EEG evidence through a small set of soft hyperedges rather than pairwise edges. A per-channel Mamba backbone produces one token per (channel, second), and a spatiotemporal hyperedge block pools these tokens into E_h shared group embeddings through soft memberships and broadcasts them back. The same encoder serves three downstream tasks: window-based detection, one-second point-wise detection, and preictal seizure prediction. On TUSZ and CHB-MIT, HyBrain achieves the best AUROC on every reported setting against ten baselines, with the largest gap on long-clip preictal prediction. It also matches the most efficient baselines in training time and peak GPU memory. A qualitative analysis shows that even a single learned hyperedge cleanly captures the preictal -> ictal -> postictal trajectory on a real seizure clip.

LLM agents performing long-horizon tasks accumulate tool results that later steps may need. Passing the full history to every invocation is costly even when it fits within the context window, while reducing it risks omitting needed information. Existing context management methods can overlook how earlier tool results are used in subsequent execution, leaving needed information out of context. We introduce ContextRender, which manages context through a persistent graph of execution dependencies. We develop Tool-Flow Analysis to track how later operations reuse information from earlier tool results, providing a signal called observed reuse. A renderer combines this signal with recency and semantic relevance to select results within a fixed history budget, retaining omitted results for later use. Across AppWorld and 8-objective QA with three execution models, ContextRender outperforms the evaluated context management baselines using a 6K history budget, well below the models' maximum context windows. Within this budget, it achieves task performance close to or above that of passing the full history while reducing mean inference cost by 10.2%-32.2% relative to Full history. Ablations show that observed reuse improves task performance and retention of results reused later.

Sparse mixture-of-experts (MoE) large language models scale model capacity by routing each token to a small subset of experts. Their routers are regularized with load balancing terms and learn affinity scores through the language-model objective. However, these objectives do not provide direct alignment between routing affinities and token-level error. We introduce token-error supervision for sparse routing in two forms. The first form predicts an error score per expert. The affinity-weighted aggregate of these scores is aligned to the next-token cross-entropy loss, while the individual scores attenuate affinity before top-$K$ selection. The second directly aligns the router's affinities to the model's objective without requiring an additional head or inference-time modification. Both formulations use the Itakura--Saito divergence or an exponential negative log-likelihood for aligning affinities and token errors. Across two sparse MoE backbones and four multiple-choice question-answering benchmarks, we evaluate both supervision mechanisms. On Granite, our method improves accuracy by approximately 2.3 percentage points on average over a parameter-matched routing baseline. With stronger supervision, the gain on ARC-Challenge reaches 2.94 points. Both mechanisms preserve the native sparse execution budget and aggregation policy. Our code is available in the supplementary materials.

Artificial Intelligence Virtual Cells (AIVCs) are envisioned as scientific agents that simulate cellular responses, explain underlying mechanisms, and support hypothesis-driven discovery. Existing AIVC benchmarks, however, operate primarily at the simulation layer, motivating complementary evaluation of how models interpret experimental evidence and formulate biological hypotheses. We introduce OmniVCBench, a figure-centric, source-traceable benchmark for the interpretation component of an AIVC. It contains 6,077 curated single- and multi-subfigure question--answer pairs derived from figures and experimental contexts in the scientific literature. Guided by Bloom's taxonomy, we instantiate interpretation-layer counterparts of the AIVC Predict--Explain--Discover agenda through three scientific reasoning tasks. We further introduce AIVC-Judge, a task-conditioned MLLM-as-a-judge framework with category-specific, reference-aware rubrics for evaluating open-ended responses. A complementary Model-Derived Hard-Negative Mining (MDHNM) strategy converts plausible errors observed during model inference into MCQ distractors for lower-cost evaluation. Within the evaluated heterogeneous model pool, MCQ accuracy correlates positively with AIVC-Judge scores, providing a complementary view of performance alongside open-response evaluation. Code and data demo are available at https://anonymous.4open.science/r/OmniVCBench.

Adam is widely observed to remain stable even when the objective deviates significantly from global smoothness. Under the generalized smoothness framework, however, existing analyses rely on strong tail assumptions on the stochastic gradients, such as almost-sure boundedness or sub-Gaussianity. Whether Adam converges on generalized smooth objectives under only second moment information on the stochastic gradients, without such concentration assumptions, was identified as an important open direction by Li et al. (2023). This paper gives an affirmative answer under fairly general conditions: such tail assumptions are not necessary. Building on the Adam self-normalization framework of Jin et al. (2026), developed for classical smoothness and bounded variance, we extend the stopping-time and de-preconditioning strategy to the $L_0$-$L_p$ generalized smoothness condition and a generalized second moment ABC condition. Even when the stochastic-gradient condition provides only second moment information that may grow along the trajectory, the stochastic trajectory of Adam remains in a locally well-behaved smoothness region, with stretched-exponential tail decay under bounded variance and global smoothness. Consequently, we establish high-probability convergence rate guarantees over the full range $p<2$, with confidence dependence of order $\delta^{-1/2}$, while the stepsize prefactor depends on $\delta$ only through a single logarithmic factor. We further construct a hard instance showing that, under only second-moment information, this $\delta^{-1/2}$-type confidence dependence is sharp. Finally, in the regime $p<1$, we combine the trajectory control with polynomial-growth estimates on rare events to obtain convergence rate guarantees in expectation.

Every day, people continuously infer situational context and adjust the way they understand and remember the world. Context, signaled by the prefrontal cortex, is known to modulate working memory and episodic memory, but the algorithmic understanding of this modulation remains limited. Here, we train a recurrent neural network (RNN), augmented with an episodic memory buffer, to infer context using Bayesian inference as it continuously makes predictions of upcoming scenes while watching naturalistic movies. When the inferred context modulates the RNN's recurrent connectivity (the basis of working memory) in a low-rank manner, the model's activity patterns best match neural responses in human participants who watched the same movies during fMRI. Context also modulates episodic memory retrieval, such that the model retrieves memories based on not only content similarity but also context similarity. This is implemented as a key-value system with self-attention, designed to additionally encode context and retrieve context-congruent memories. The resulting model not only better resembles human brain representations but also learns to retrieve memories like humans much faster than a model without context modulation. Together, our findings suggest a computational mechanism by which context modulates information maintenance and long-term memory retrieval in naturalistic environments.

Video diffusion transformers (DiTs) increasingly adopt mixture-of-experts (MoE) architectures to reduce active computation, but their full expert storage remains costly. Existing one-shot pruning criteria mainly rely on static activation or routing statistics and cannot capture layer-level re-routing after expert deletion. We introduce DIET, a training-free expert pruning framework based on deletion responses. A single all-expert calibration pass records expert outputs and router states for matched conditional and unconditional tokens. Candidate deletions are then replayed from cached tensors, requiring no additional model forward passes. The resulting deletion-response signatures characterize each expert by the changes induced when it is removed. DIET selects retained experts by minimizing Overall Diversity Loss (ODL), which preserves directional coverage in signature space, and combines intra-layer local search with an inter-layer regression-guided budget search to allocate experts across layers. On LingBot-Video 30B-A3B, pruning 50% of experts (6,144 to 3,072) reduces the checkpoint from 57 GB to 30 GB and enables single-card deployment on a 48 GB GPU without fine-tuning. Under a fixed 284-case VBench protocol, the VBench Total increases from 0.7941 to 0.8115. Across tested retention budgets, DIET consistently outperforms competitive pruning baselines adapted from large language models.

Certo is a small non-generative decision model (Qwen3-4B): it scores candidate actions from their text and returns a probability, instead of generating an answer. The accurate design reads the state, the rules, and each candidate together (a joint scorer), so cost grows with the menu. Independent encoding lets each candidate be encoded once and reused across states (about 5x cheaper at 77 candidates), but separates state from candidate. We ask how much rule-sensitivity survives that move, and whether it can be trained back. Four experiments on Certo: (1) the tested conversion to cacheable scoring loses rule-sensitivity (recall@1 1.00 -> 0.24) while the joint scorer holds 1.00, and a shortlist+rerank rescue fails; (2) targeted counterfactual supervision restores strong performance on held-out synthetic rule tasks (paraphrase, counterfactual, composition; reproducible across seeds), though we do not isolate whether predictions depend on the supplied rule; (3) on real rules the added benefit is not established -- after fixing a truncation confound, the joint scorer wins significantly on the short tier (0.861 vs 0.500) and directionally on the hard tier (0.655 vs 0.483, n=29); (4) a matched cross-domain real-prose mixture did not help and reduced contract accuracy (-9.3, -16.2 points). A cacheable encoder can be made rule-sensitive on its training distribution, but transfer to unseen-source real rules is not established; the joint scorer keeps an edge at the cost of caching.

Harness optimization provides a practical setting for recursive self-improvement (RSI), where agent-generated modifications inform subsequent changes through execution feedback. Recent work such as Meta-Harness implements this process through iterative code generation and evaluation, but retains a fixed development set and proposal policy. These constraints channel evolution along a single search trajectory, increasing the risk of converging to a local optimum. We make the improvement process itself adaptive by organizing search into branches with evolving development subsets and proposal policies. Each branch retains development cases solved by more of its leading harnesses than by those of other branches, drops cases solved by every leading harness across all branches, and revises its proposal policy using its own search history. To deploy the resulting complementary harnesses, we propose a router to select one development-selected branch head for each new input before execution. Across mathematical reasoning and agentic coding benchmarks, our system achieves relative improvements over Meta-Harness of 34.8% on Olympiad-level mathematical reasoning, 11.6% on Terminal-Bench 2.0, and 3.8% on SWE-bench Lite, with harness selection and router configuration based solely on development data. These results show that evolving branch objectives and proposal policies can yield complementary harnesses whose strengths a router combines without access to test outcomes.

Sparse autoencoders (SAEs) are increasingly scaled to wider dictionaries to recover fine-grained structure from large language model activations. However, a feature is useful for interpretation only if it remains a stable unit of analysis when the same meaning is expressed in different surface forms. We study this reliability question for TopK SAEs via feature sensitivity. Experiments demonstrate that scaling selectively reduces the sensitivity of rare features, while common features remain comparatively stable. A controlled width\(\times k\) factorial experiment identifies the active budget k as the root cause: the degradation arises from the selection boundary rather than dictionary width alone. We attribute this failure to the geometry of TopK selection. The active margin, the distance to the cutoff, predicts feature loss without thresholds. Guided by this margin diagnosis, we introduce pairwise rank stabilization. Our method targets ordering failures at the cutoff and improves rare-feature sensitivity by \(8.83\) percentage points, while keeping reconstruction and alive-feature coverage near the baseline. Overall, our results suggest that wide TopK SAEs should be evaluated not only by reconstruction, sparsity, and feature count, but also by feature reliability under semantic variation and boundary geometry for stable interpretability.

Inverse design of physical systems (molecules, devices, microstructures) often reduces to optimizing a high-dimensional structure against an expensive black-box simulator. Direct search is difficult because the space is non-Euclidean, feasibility is hard to encode, and each evaluation is expensive. We present Co-PiLOT, a latent optimization approach that maps candidates through a generative encoder-decoder, uses the decoder as a learned validity prior, and searches the latent space with physics-informed black-box optimization. The framework is applied on the inverse design of magnesium alloy microstructure/texture. We develop a vision transformer based-encoder; paired with latent diffusion, diffusion transformer and rectified-flow transformer-based decoders on $\sim80{,}000$ EBSD-derived microstructure dataset to learn a minimal bottleneck, $z$. The ViT-FMDiT model ($z$=$768$) reconstructs high-fidelity microstructure images (FID $27.86$, MS-SSIM $0.178$), which our self-segmenting orientation codec converts into input grids for crystal plasticity solver. Finally, we introduce MERIDIAN, an active latent optimizer driven by deep-kernel Gaussian-process uncertainty, failure-aware feasibility prediction, manifold-aware trust regions, and target-aware acquisition. Within a budget of $160$ simulations, the ViT-FMDiT and MERIDIAN combination yields the best target-driven objective score, reducing the relative target error by $3$--$22\%$ against seven baselines (DANTE, TuRBO, BAxUS, CMA-ES, DDOM, SEIKO, DDPO) on the same decoder.

Reinforcement learning for long-horizon agents typically relies on sparse outcome-based rewards. This leads to a severe cold-start problem, as early-stage policies often fail to solve sampled tasks, leaving little useful reward signal for learning. To mitigate this problem, we use on-policy distillation (OPD) to provide token-level guidance on the student's own rollouts. We find that the benefit of this guidance depends on the performance gap between the teacher and the student. When the teacher substantially outperforms the student, distillation helps guide the student through the early training stage where outcome rewards provide little learning signal. As the gap narrows and eventually reverses, however, continued distillation becomes less beneficial and may hinder further improvement. Motivated by this observation, we propose Gap-Adaptive Teacher Scheduling (GATS), which augments the student's RL objective with an OPD term whose weight adapts to the teacher-student performance gap. Specifically, GATS gradually reduces teacher guidance as the student approaches the teacher's reference performance and withdraws it once that reference is reached. This enables GATS to leverage task-trained teachers smaller than the student, since teacher guidance is primarily needed during early training. Across ALFWorld, WebShop, and ScienceWorld with three Qwen2.5 teacher-student configurations, GATS achieves the highest average success rate among the compared methods in all three configurations, improving over reward-only GRPO by 4.37%-11.87% under matched student rollout budgets. Code is available at https://github.com/Ricardo-H/guide-then-let-go.

Existing LLM routers choose among models using static per-model costs. We show that open-weight inference markets introduce a second, largely ignored decision axis: after choosing a model, a client must still choose which provider serves it. Measuring live endpoints across [nummodels] open models, competing providers, multiple task types, and three measurement waves, we find that provider choice cannot be inferred from the price list. The same model can vary sharply in quality, latency, availability, and price across providers; higher-priced providers are consistently faster, but price does not reliably predict quality or availability; and provider feasibility is task-selective, with one deployment nearly normal on knowledge tasks but catastrophically degraded on multi-step reasoning. We formulate same-model provider selection as a price-taker market-aware routing problem. A simple measured-map policy routes to the cheapest provider that is both quality-equivalent and healthy, yielding matched-quality savings while avoiding degraded endpoints. Because the map drifts, we introduce FACET, an online provider router that certifies per-(provider x task) feasibility facets and fails safe to an anchor before serving uncertified arms. Across relaxed deployment assumptions, FACET tolerates imperfect task assignment and sparse feedback, while systematic evaluator bias exposes a quality-signal trust boundary that can be mitigated with ground-truth probes or audits. Live provider runs further confirm that certification can move real traffic from a premium anchor to a substantially cheaper certified endpoint. Our results suggest that market-aware LLM routing must measure not only which model to use, but also who serves it.

Forecasting time-varying functional connectivity from electroencephalography (EEG) requires modeling both history-dependent trends and structured variability across channels. Conditional flow matching provides a framework for distributional forecasting, yet it remains unclear whether graph-informed source distributions offer practical advantages over isotropic noise and strong deterministic predictors. We introduce a graph-structured residual flow framework that separates conditional mean prediction from stochastic residual transport. A history-only predictor estimates the future connectivity graph, while a graph Gaussian source encodes dependencies derived from past connectivity through a Laplacian-based covariance. A conditional velocity field transports source samples to future graph residuals, with transport time explicitly distinguished from physical EEG time. Our study identifies the conditions and controls needed to distinguish useful residual transport from improvements attributable to deterministic prediction, learned representations, and sampling effects.

Complex tasks inherently couple workflow structure, agent responsibility, collaboration, and memory access: task regions delimit responsibility and tool scope, cross-region dependencies give rise to handoffs, and ownership boundaries delimit private and selectively shared memory. Existing methods can jointly optimize agent and communication structures, yet such optimization does not by itself require responsibility, handoff, and memory boundaries to remain consistent with task dependencies. We term this requirement topological coherence. We introduce TOCOMAS, a Topology-Coherent Multi-Agent System. TOCOMAS grounds a task graph in tool interfaces, organizes compatible task nodes into reusable responsibility domains, and derives dependency-induced and profile-conditioned collaboration together with boundary-regulated memory visibility. During online self-evolution, TOCOMAS proposes coupled changes to agent, collaboration, and memory policies, retaining for subsequent tasks only candidates that satisfy structural constraints and improve evaluated reward. Across BBEH, WorkBench, SWE-Bench-Verified, and CoMemBench, TOCOMAS improves task success over baselines across backbones. CoMemBench also shows gains over the self-evolving baseline in verified progress, handoffs, and memory isolation.

Brain networks characterize structural and functional relationships among brain regions and support research on cognition, brain disorders, and brain-computer interfaces. Their time-varying topology and higher-order spatiotemporal dependencies are not adequately represented by conventional static networks. Existing tools primarily focus on static connectomes and provide limited integration of dynamic network modeling with modern graph and sequence learning methods. We present BrainNet Studio, an integrated toolkit for static and dynamic brain network analysis. It provides a unified workflow encompassing network construction, feature extraction, predictive modeling, candidate biomarker identification, visualization, and assisted interpretation. The toolkit integrates 27 algorithms, including deep learning, graph neural networks, and spatiotemporal sequence models, to support classification and the identification of discriminative brain regions and connections. A large language model generates researcher-verifiable summaries of functional connectivity, structural connectivity, and structure-function coupling at individual and group levels. Within a consistent computational framework, users can configure analytical tasks, compare methods, inspect outputs, and extend functionality without repeatedly assembling application-specific pipelines. BrainNet Studio provides a practical and extensible platform for connectome analysis in cognitive neuroscience, exploratory studies of brain disorders, and brain-computer interfaces. The toolkit is publicly available at https://github.com/xbrainnet/Brainnet-Studio.

Advances in the coding capabilities of LLM agents allow them to inspect and modify their own instructions, tools, and execution procedures. Existing approaches use this ability to search for improved agents through repeated downstream evaluation, which incurs substantial costs and ties the search to the evaluated tasks. We introduce \textbf{SelfSearch}, a reward-free search procedure in which agents modify themselves using records of previous self-improvement episodes. These records capture the reasoning, tool actions, and outcomes of earlier modification attempts, providing concrete experience for improving both task solving and self-modification. Without downstream reward signals during search, SelfSearch improves population-mean success over the initial agent in all six model--benchmark settings, with individual agents gaining up to 11.2 percentage points on Terminal-Bench 2.1. On SWE-bench Multilingual, an agent improves success by \textbf{5.0} percentage points while reducing execution cost by \textbf{38.5}\% on tasks solved by both the initial and evolved agents. SelfSearch achieves competitive task success with evaluation-guided search baselines at lower search cost. With only \textbf{\$4.03} in search cost, it produces a harness that solves \textbf{82.0}\% of Terminal-Bench 2.1 tasks with DeepSeek V4 Flash under the settings of a public nine-harness comparison, matching the top-scoring harness, Codex. These results suggest that experience gained through self-modification can improve agents' downstream capabilities and efficiency.

As the context size of text processed with an LLM grows, the size of KV caches can outstrip the memory allocated for the original model weights. This impacts LLM throughput negatively, since decoding is memory-bound and decode cost grows with cache size. Recent work alleviates this bottleneck by discarding the least relevant tokens. Eviction introduces a tension, since a one-off decision to discard content may prove detrimental later. Instead, we focus on alternative choices that can lead to cache compression without evicting tokens. We achieve this by learning a selector that is able to produce, based on context, a per-layer cache configuration towards an overall compression budget. The selector operates along three axes: sharing one cache across layers (depth), caching at fewer bits (precision), or truncating the low-rank latent cache representations (rank). We call the resulting method KV-Kaizen, for the many small per-layer choices it compounds. We observe that these interventions taken independently and uniformly over all layers limit achievable compression because they degrade accuracy. Crucially, composing them locally and adaptively to the context can instead preserve accuracy while achieving large memory savings. At inference, the selector runs once, before pre-fill. In evaluations on instruction following and reasoning tasks, our selectors reach the Pareto frontier of accuracy against cache size, against learning-free and post-hoc baselines. On long-context tasks, KV-Kaizen improves on eviction and can be composed with it, reaching a 32x smaller decode-time cache on a 14B model while preserving accuracy. A 4x cache size reduction incurs no accuracy degradation from 7B parameters up, and a compressed model is more accurate than a smaller uncompressed one with the same cache size. Together, these findings support pre-training large models and compressing them only afterwards.

Brain-AI alignment is often interpreted as a sign that model and brain perform similar computations. Whether the aligned units are causally involved in model computation is rarely checked. On an abstract pattern-completion task (AAABAAA $\rightarrow$ B), we compare LLM attention-head representations with human EEG and test how ablating those heads affects task performance. Alignment and causation dissociate: brain-aligned heads contribute to performance, but their removal is substantially less disruptive than removal of heads selected via attribution patching. We compare two head sets that prior interpretability work defines without reference to the brain: concept vectors (CVs), which represent abstract patterns across formats, and function vectors (FVs), selected for their contribution to correct-answer prediction. Brain alignment shows little association with FV scores, while its association with CV scores varies across models. Among brain-aligned heads, we find recurring attention profiles: one emphasizes distinctive elements (novelty heads), the other repeating elements (repetition heads). The novelty family tracks salience and attends to the same elements that humans look at, yet its removal is less damaging than random ablation on average. Repetition heads contribute modestly to performance and are associated with abstract-pattern representation (CVs). Across 17 models spanning 3B-72B parameters, FV-ranked removal is substantially more disruptive than brain-ranked removal. Brain alignment thus captures how the model reads the stimulus, and only faintly captures how it represents the pattern and solves the task.

Large language models (LLMs) are increasingly proposed as decision assistants who must reason probabilistically from available evidence under explicit decision costs. We propose a decision-theoretic framework that decomposes LLMs' decision loss into two components: forming accurate beliefs from provided evidence and translating those beliefs into actions that optimize a provided utility function. Using a synthetic benchmark with known ground truth, we apply the decomposition to characterize probabilistic reasoning in frontier and open-sourced models. We further evaluate whether RL interventions targeting beliefs, decisions, or both improve these components across three domains, whether improvements transfer across components and elicitation formats, and whether decision performance can improve without improvement in belief formation. We find that targeting one component of probabilistic reasoning redistributes decision loss, improving the target without necessarily transferring to others, and that jointly targeting belief formation and decision-making improves both but hinges on matched formats between training and evaluation.

Running a language model locally offers advantages in privacy, latency, and cost, but local hardware fits only small models, which are less capable than frontier models. The usual remedy for a hard query, escalating it to a cloud model, gives up the privacy and cost advantages of running locally. A deployment that stays local faces two decisions for hard queries instead. First, it can spend more computation on a query, e.g., reasoning before answering, which raises accuracy at a cost in latency, so it must decide which queries are worth the extra computation (efficiency). Second, some queries are beyond the local model, and delivering a wrong answer is worse than deferring the query to a human in the loop, so it must decide which answers are safe to deliver (safety). We show that both decisions can be made from the model's own hidden states. The prefill state, computed before any token is generated, predicts whether the model will answer correctly, and the answer state, at the end of the generated answer, predicts whether that answer is correct. HARISSA fine-tunes the model so that both states predict correctness, then makes both decisions with one policy that cascades through the ways of answering from cheapest to most expensive, skipping a way the prefill state predicts will fail and deferring the query when the answer it stops with is predicted wrong. On a device running a single model, HARISSA is within one accuracy point of chain-of-thought at 2.7 times lower latency. On a server holding four sizes of one model, HARISSA is more accurate than the FrugalGPT and Self-REF cascades at the same latency, and at the same deferral rate the answer state leaves fewer wrong answers than the standard confidence signals in five of six task and setting pairs.

Physics-Informed Neural Networks (PINNs) embed governing equations into deep learning, but enforce them only through loss residuals, leaving highly oscillatory wave behavior to be discovered by optimization. As a result, methods that achieve relative $L_2$ errors below $10^{-3}$ on standard manufactured Helmholtz benchmarks can fail on practical radiation problems involving singular excitations, absorbing boundaries, and wave fields spanning tens of wavelengths. Architectural physics embedding addresses this limitation by factorizing the field into analytically derived oscillatory kernels and learnable envelopes. However, the kernel dictionary must be manually constructed and scales with the number of elementary units, growing exponentially with the depth of hierarchically structured systems such as antenna arrays and metasurfaces. We propose PE-EK-PINN (Physics Embedded with Evolving Kernels), which treats physics kernels as reusable learned representations rather than fixed analytical inputs. A converged subsystem field is frozen and promoted to an evolved kernel, whose transformed copies are reused to represent higher-level configurations without deriving new governing equations. The resulting hierarchy makes the peak number of active kernels independent of system size and reduces cumulative training cost from $O(N)$ to $O(\log N)$. Experiments on dipole arrays, composite line-source geometries, and cross arrays demonstrate the dramatic training cost reduction, while achieving a reduced or comparable relative $L_2$ error. One notable example is PE-EK-PINN solves a $256$-dipole array more than 30 times faster than direct PE-PINN.

An agent skill is a reusable, actionable natural-language artifact that guides an agent to perform a task effectively under a given harness. Recent studies have explored the optimization of agent skills, contributing to a growing collection of publicly available skills spanning diverse tasks, domains, and harnesses. Despite millions of publicly shared skills, existing skill optimization methods largely overlook this accumulated knowledge, instead relying solely on expensive agent rollouts to iteratively refine skills for a target task. To address this, we propose \textbf{Retrieval-Augmented Skill Optimization (RASO)}, a framework that leverages an external skill corpus as prior knowledge throughout skill optimization. RASO retrieves relevant knowledge from existing skills and adapts it to the target task and harness via Cross-Harness Adaptation, accounting for mismatches in both domain and harness. RASO comprises two complementary stages: \textbf{Retrieval-Augmented Skill Initialization (RASI)} constructs a knowledge-grounded initial skill without requiring agent rollouts, while \textbf{Retrieval-Augmented Skill Update (RASU)} iteratively refines the skill by retrieving external knowledge guided by execution feedback. Across four agent benchmarks and two models, extensive experiments show that RASO consistently outperforms baselines without retrieval-augmented skill initialization and updating.

Interactive agent benchmarks and multi-turn reinforcement learning increasingly place a second language model in the role of the user. This simulated user controls what information the agent receives and when, yet current benchmarks score only the agent and do not directly measure whether the user correctly executed its assigned role. We introduce UserProxyBench, an evaluation layer over the tau-bench family, and the User Fidelity Score (UFS), which measures adherence to the benchmark's private user instructions using task-grounded rubric criteria scored independently of agent success. Holding the agent fixed at GPT-5.5 and varying only the user proxy across 375 enterprise tasks changes mean task reward by 15.2 points, while 24.4% of successful episodes contain a user-specification violation. The dominant failure is premature disclosure: users provide information before it is requested. This behavior has little effect on task reward, yet among successful episodes it causes the agent to make 1.06 fewer tool calls on average, changing the interaction being evaluated while preserving the reward. Finally, across seven proxies we identify an empirical cost-fidelity frontier, enabling practitioners to select the least expensive simulator that satisfies a required fidelity level.

As the chain-of-thought reasoning capabilities of large language models improve, evaluating and calibrating their reasoning confidence is becoming increasingly important for quantifying the uncertainty of their answers. Current methods for estimating the confidence of large language models are generally based on probabilities of selected key tokens, but the underlying mechanism remains unclear. Our pilot study finds that replacing selected token probabilities with coarse substitutes can also improve calibration, motivating us to further explore effective signals of model confidence. We introduce Divergent Token Confidence (DTC), a framework that estimates confidence by counting tokens at which two models strongly disagree during decoding. DTC identifies these divergent tokens using the Jensen-Shannon divergence between next-token distributions evaluated along the same reasoning trajectory. We find that their count is almost negatively associated with answer accuracy, thereby serving as a simple yet effective signal for uncertainty quantification. DTC supports both white-box and black-box evaluation using auxiliary models, without explicit training and affecting the generation process. Experiments across multiple model families and six mathematical benchmarks demonstrate improved calibration over probability-based and verbalized baselines. Under white-box evaluation, the count-only estimator achieves an average expected calibration error of 13.0%, compared with 32.7%-42.4% for standard full-sequence confidence methods. In black-box settings, it also improves calibration over the original verbalized scores. For example, mean expected calibration error falls from 32.1%-40.2% to 13.7%-16.3% on DeepSeek-V3.2. These findings provide new insights for improving reasoning uncertainty quantification in large language models. The code is released at https://github.com/szu-tera/DTC.git.

Risk aversion in resources could prevent misaligned AI agents from causing catastrophic harm. Misaligned but risk-averse agents would tend to favor safer strategies like making deals with humans over riskier strategies like rebelling. We train agents to be risk averse through character training, finding that persona traits provide a robust mechanism for instilling risk preferences. To do this, we construct a model constitution describing constant absolute risk aversion (CARA) over an agent's resources and instill it through on-policy distillation. Despite never seeing the benchmark's decision format during training, character-trained models are competitive with baselines trained directly on it, and generalise better than them out of distribution on two of our four models. We also modulate different aspects of the constitution, finding that token budget and model choice are the most influential aspect of character training to instill risk aversion. We conclude from these results that character training is a promising and scalable way to instil broad dispositions, which we can use to our advantage in mitigating risk from misaligned AI agents.

Current vision-language models (VLMs) encode visual information in dense hidden states where object identity, spatial layout, and local attributes are implicitly entangled rather than explicitly disentangled, limiting their ability to isolate and modulate the specific visual evidence required by a given language query. Inspired by sparse population coding and top-down modulation in biological vision, we introduce NeuronEye, a plug-in framework that constructs a sparse, concept-level neuron vocabulary from intermediate VLM representations and selectively activates query-relevant visual concepts during inference. NeuronEye decomposes vision-token states into an overcomplete sparse basis organized by concept-level clusters, uses the language query to activate relevant clusters and localize the patches where selected concepts are expressed, and injects the focused evidence back into vision tokens. A complementary suppression mechanism attenuates dominant perceptual directions to preserve weaker but relevant cues. All operations run in a single forward pass over a frozen VLM backbone. On Qwen2.5-VL-7B, NeuronEye raises CV-Bench overall accuracy by +3.1 with gains of +9.5 on Distance, and improves BLINK Multi-view by +8.3, with similar trends on LLaVA-1.6-7B. These results suggest that sparse neuron vocabularies can serve not only as post-hoc interpretability tools but also as active interfaces for concept-level visual reasoning.

Large language models (LLMs) enable agents to solve long-horizon tasks by generating a plan and then executing it in an environment. However, successful planning requires two distinct capabilities: selecting an appropriate plan for the task and executing it faithfully. Existing planner--executor systems can fail at either stage, while final task success alone cannot distinguish selection from execution failures. We therefore study the Plan Declaration--Execution Gap and introduce Planning-as-Routing, where an LLM declares one of four planning modes: Predefined, Sequential, Hierarchical, or Search, and a deterministic router dispatches the task to the corresponding pattern-specific executor. Across four benchmarks and three LLMs, we find three consistent patterns. First, generic Plan+ReAct often fails to preserve declared planning structure, especially for longer plans: across three benchmarks, only (22)--(45%) of trajectories preserve it, whereas pattern-specific executors enforce the intended structure. Second, planning-mode effectiveness varies across environments and models: Search performs best on ALFWorld, Hierarchical on SWE-bench, and the strongest pattern can vary across models within the same benchmark. Third, the largest gains come from execution: pattern-specific executors improve task success from (0.48) to (0.92) on ALFWorld and from (0.36) to (0.44) on SWE-bench Verified over Plan+ReAct. Current LLMs, however, do not reliably select the strongest mode for each task, although few-shot examples improve selection in some benchmark--model combinations. Overall, reliable agent planning requires both effective mode selection and faithful execution: routing substantially closes the execution gap, while task-specific mode selection remains open.

Verifying a vision-based neural feedback system requires a model of the observations its controller acts upon. Such a model must capture the variation the sensor produces, while remaining tractable for closed-loop analysis. Generative adversarial networks (GANs) have served as perception surrogates, but they are large, reproduce complex scenes poorly, and are hard to verify. We explore stochastic world models as a richer class of perception surrogates. We train a world model with physically grounded latents, built from operations that standard verifiers bound. It reproduces held-out frames more faithfully than GAN surrogates with up to 130 times as many parameters. To verify these surrogates, we develop a procedure that combines falsification, adaptive refinement, symbolic, and backward analyses. On an emergency braking benchmark with a GAN surrogate, our procedure resolves the entire state space, 38% of which the state-of-the-art verifier left unresolved. On the RGB version of the benchmark, where no verification results have previously been reported, our procedure resolves over 80% of the state space with a world model surrogate.

A small trainable advisor can steer a frozen language-model executor using natural-language advice. In addition to learning from task rewards, the advisor can use feedback from completed interactions to improve its advice. However, a plausible correction need not change execution, yet learning from such corrections can still affect the advisor's future decisions in other contexts. In a shared-parameter model, we prove that such corrections can limit learning if their targets favor useful advice less strongly than those of other corrections. Keeping them less often than the rest improves the model's eventual performance compared to learning from every correction. Motivated by this, our method, Advisor Self-Distillation (AdviSD), pairs outcome-based reinforcement learning with self-distillation from a feedback-conditioned copy of the advisor selectively. Reflection proposes corrections, and the advisor scores the same recorded executor response with and without its issued advice, using the magnitude of the difference to select decisions for supervision. This approach does not require executor likelihoods or additional executor rollouts. Experiments with Qwen3-8B advisors for Gemini and Claude show that AdviSD outperforms advisor-GRPO by 4.2-6.4 percentage points on BFCL-v3 and by 3.9-5.1 score points on EnvScaler. The trained advisors generalize to out-of-domain tasks and transfer across different executor versions and model families. AdviSD also beats matched-count random selection, supporting the value of its selection rule.

Agent performance depends on both reasoning ability and the environment in which it acts. We study test-time AI-for-AI, asking how a Builder can learn to construct better execution environments for a Target while both models' weights remain fixed. To make the Builder's experience reusable, we introduce Meta-Skill: principles specifying when support is needed and what resources to provide. The Builder learns these principles from Target's execution feedback on the development set, then uses the frozen skill bank to construct harnesses for unseen tasks. Across Harness-Bench and NewtonBench, full-bank meta-skills improve macro-average performance by 8.95 percentage points over no-skill construction, and 12.02 points over direct delivery of the same bank to the Target. These results highlight the value of translating experience into executable support. Gains when the same model serves both roles further suggest a path to system level self-improvement through learning to build better environments.

As agents take on longer and more complex problems, controlling the execution becomes a task in its own right. Each step in the run brings new control choices, like which partial work to build on, whether to start fresh, or when to stop. We introduce agentic meta-reasoning, an inference-time harness that makes these choices an explicit and structured reasoning process. Workers carry out the task-level computation, while a controller consolidates what the run has established, explores next options, assesses what each option is worth under the remaining budget, and dispatches the chosen work with context drawn from persistent memory. Between decisions the controller carries only a compact account of the run rather than replaying its full history. Our baselines span production coding agents and research harnesses, together with a Direct Control Agent using the same workers and compute budget allowance. On ProgramBench, which tests long-horizon agentic capability through program reconstruction, meta-reasoning achieves 71.5% with GPT-5.5 against 58.0% for Codex; with Opus 4.8 it achieves 67.2% against 65.5% for Claude Code. On the other benchmarks, spanning abstract reasoning, multi-domain long-horizon reasoning, and proof generation, it gains between 3.6 and 4.2 points over direct control, averaged across three frontier models. It keeps improving over the tested budget ranges where direct control plateaus, though its overhead can hurt at small budgets. Artifact-graph analysis reveals more reuse of earlier work, higher coverage of correct solutions in most settings, and nonuniform gains in final selection. These results indicate that spending computation on structured control becomes more important as agents scale to longer runs.

Effective regulation of AI is a defining policy challenge, driven by their integration into all aspects of society. To remain responsive to their rapid development and emergent properties, policymakers across the globe rely on high-level principles and abstract legal requirements. Yet, while this flexibility supports future-proofing human-centred regulations and aligning them with socio-ethical values, it also causes legal uncertainty downstream as developers, companies, and auditors struggle with translating these abstract requirements into verifiable technical requirements. Using the AI Act as an example, this paper draws on Coleman's bathtub to analyse the regulatory learning space in AI governance. It argues that legal uncertainty cannot be fully reduced ex ante and that, within reasonable bounds, it is also necessary for regulatory learning because it creates the space in which boundary negotiation over socio-technical meaning can occur. Building on this analysis, the paper shows how boundary objects and boundary negotiating artifacts help explain the translation of legal requirements into operational practice. By examining technical sandbox frameworks, it further identifies concrete properties that technical infrastructures must possess to function effectively as boundary negotiation artifacts in AI assessment. The paper concludes that legal certainty remains the long-term aim, but that premature closure of regulatory instruments risks undermining the learning processes needed for adaptive governance.

While neural theorem provers have achieved impressive milestones in formal mathematics, they largely operate on the assumption that faithful Lean 4 formal statements are already provided. Translating informal natural language into a formal language is a critical data bottleneck plagued by an "illusion of rigor": standard type-checkers accept statements that compile but drop hypotheses, introduce vacuous truths, or subtly alter mathematical bounds. To resolve this, we introduce Sage (Semantic Agent-Guided Formalization Engine), an agentic framework that replaces monolithic translation with a four-stage decomposed generation pipeline coupled with a dual-signal semantic correction loop. By pairing Lean 4 compiler diagnostics with multi-dimensional semantic feedback, our correction loop enforces mathematical fidelity alongside syntactic validity. By explicitly accounting for the gap between open-ended queries and declarative formal targets, our pipeline prevents models from achieving high formalization rates by guessing unverified answers (exhibiting a 70.9% answer leakage rate). Consequently, Sage suppresses leakage to 2.7% while achieving 73.3% pass@4 joint compilation and semantic fidelity on the Omni-MATH without proofs (compared to 42.0% for a fine-tuned Goedel-Formalizer-V2 baseline). Finally, on IMO-Unformalized, a novel frontier of 175 unformalized International Mathematical Olympiad problems, Sage demonstrates effective zero-shot generalization with 87.4% pass@4 verified fidelity compared to just 19.4% for the baseline, winning over 79% of blind pairwise evaluations.

Asthma deterioration forecasting must remain reli- able when patient populations, sensor ecosystems, and available modalities change across cohorts. Existing models commonly optimize within-cohort discrimination and may produce poorly calibrated probabilities after transfer. We present CALIBRA, a calibration-first multimodal temporal framework for short- horizon risk prediction with incomplete data. Dedicated recurrent encoders process environmental, pulmonary, symptom, medication, wearable, and context streams; a reliability-conditioned gate suppresses stale or absent modalities, while gradient-reversal training discourages avoidable cohort signatures. A shrinkage- based hierarchical logistic layer calibrates probabilities using a patient-disjoint target subset, and split conformal prediction provides abstention-capable prediction sets. To avoid fabricating clinical evidence, we evaluate the complete implementation on a documented three-cohort semi-synthetic benchmark with controlled distribution shift, informative missingness, and sealed target patients. Across five configured seeds, CALIBRA achieved mean target-test AUPRC 0.224 versus 0.240 for the strongest non-ablation comparator, TemporalTransformer; mean AUROC was 0.717, and Brier score was 0.098. Experiments additionally assess complete-modality failures, calibration, conformal coverage, decision curves, subgroup behavior, ablations, runtime, and parameter count. The results verify the method and reproducible pipeline under controlled shift, but do not establish clinical effectiveness. External validation on harmonized real asthma. Overall this artifact provides evidence for carefully governed real-cohort validation.

Large language model (LLM) judges are often used as rewards to train policies on objectives that deterministic verifiers cannot capture. However, these rewards are often collapsed to pass/fail ({0, 1}), which reports the verdict but not how well a response met each criterion. We model each pass/fail verdict as a score on an unreported scale, compared with one cutoff. A stretch of that scale moves every score proportionally toward or away from the cutoff, but never across it, so no verdict changes. A policy is therefore free to apply any stretch without changing anything the panel reports. Under a joint-Gaussian model, a third grade adds a second threshold and removes this affine stretch ambiguity. On MATH and SciBench outputs from one seven-criterion judge, all 14 constructed criterionwise stretches were invisible after binarization but visible with three grades. At $n=1{,}024$, a test given both population laws had at least 96.5% power at a $1.5\times$ stress. Retaining grades closes one blind spot created by binarization, but verdicts alone remain insufficient as some changes are still indistinguishable from genuine improvement. These include arbitrary within-grade changes and fixed-covariance, loading-aligned mean shifts -- the signature of a sycophancy-shaped lift the panel reads as competence. The shared-factor reference approximation fit MATH and SciBench but not HealthBench, delineating its empirical scope. We recommend keeping at least three grades (for example, asking the judge whether each criterion is fully, partially, or not met and rewarding {0, 0.5, 1}), and externally validating gains along the remaining direction, which no finer scale removes.

Large language models (LLMs) have shown remarkable capabilities in various natural language processing tasks, leading to their widespread deployment as intelligent assistants in decision-making contexts. However, the increasing complexity of these models raises concerns about their reliability, particularly regarding bias and hallucination. In this work, we evaluate the robustness of LLMs to perturbed variations of the original inquiry in decision-making tasks. We show that contrary to previous studies, perturbations can mitigate bias and hallucination in some LLMs over other models. It's found that Claude 3 is more effective for the tasks represented in most datasets, whereas models like GPT3.5 exhibit varying levels of adequacy, performing comparably in some cases but falling significantly behind in others. These insights are crucial for understanding the practical implications of deploying LLM-based assistants as effective decision-support tools in real-world applications, emphasising the need for rigorous testing and validation to ensure reliability and effectiveness. This study contributes to the growing body of research on LLM evaluation and provides insights for developing more robust and trustworthy AI assistants in critical decision-making contexts.

Training a language model on data with a narrow flaw can sometimes make the model broadly misaligned. Inspecting the data at face value often does not settle whether it will emerge, and today it is caught only after training, by auditing the resulting model. To complement post-hoc audits, we introduce Alignment Forecasting: the task of predicting alignment failures before training. Given a target model, a fine-tuning dataset, and a failure mode such as deception or sycophancy, a forecaster outputs the probability that fine-tuning would meaningfully increase that failure mode. To measure progress on alignment forecasting, we introduce ALIGNMENTFORECASTBENCH, a benchmark of over 5,000 forecasting questions spanning 17 target models, 32 datasets, and 16 failure modes. Frontier models prompted directly perform poorly on ALIGNMENTFORECASTBENCH. We therefore propose a forecasting scaffold in which an LLM reads the dataset and rates how strongly and broadly it pushes the model toward misbehavior, and a simple learned model combines that rating with the failure mode's base rate and the target model's prior tendency. This forecasts well above chance, and beats a model fine-tuned on the task and a simple forecaster allowed to see how weaker models behaved after fine-tuning on the same data. Its signals also flag problematic training examples that a frontier-model classifier misses. Filtering those examples out from real post-training data such as UltraChat results in more aligned models on our multiple-choice evaluation in most cases, though the benefit in open-ended conversations is unclear. More progress is needed before forecasts can reliably guide training data curation in practice, but our results suggest that forecasting many alignment failures before training can be tractable in the SFT setting.

Automated skill extraction underpins workforce planning, yet most systems represent skills as flat labels with no notion of the responsibility level at which a skill is practiced. The Skills Framework for the Information Age (SFIA) captures exactly this dimension, defining 147 professional skills across seven responsibility levels, but no automated LLM-based extraction targeting SFIA has been reported. We formalize the task as structured prediction of (skill, level) pairs from free text and ask three questions: how accurately can text be mapped onto SFIA's closed vocabulary, which strategies reliably predict the level alongside the skill, and do agentic designs improve on simpler retrieval and prompting? We evaluate five strategies (a lexical baseline, dense retrieval with LLM reranking, a zero-shot schema-constrained LLM, single-agent agentic RAG, and a three-agent retriever--matcher--verifier crew) against expert-mapped European ICT role profiles, all drawing on an SFIA~9 corpus built by a fully automated agentic pipeline that we release. Retrieval-based matching identifies the most skills while generative strategies are markedly more precise; only strategies assigning the level as an explicit decision predict it reliably, with similarity-based selection more than twice as inaccurate; and the crew doubles latency without improving accuracy, so added agent roles do not automatically benefit closed-taxonomy matching. These results provide the first reproducible baseline for structured, level-aware skill extraction against SFIA.

LLM agents can make unsafe tool calls even when instructed to behave safely. Existing defenses constrain agents before execution, modify tool inputs/outputs, or rely on LLM judges; these approaches may depend on model behavior or block unsafe actions without helping the agent recover. We argue that the execution environment should instead enforce safety as the agent runs and steer it toward safe alternatives when violations occur---we call this Environment Steering. We implement this by modeling the agent and harness execution state as database tables, track the record-level data flows, and check these data flows against declarative policies during runtime. When violations are detected, policy- and context-specific feedback steers the agent toward safe trajectories. On AgentDyn, this enables the agent to improve task success rate over no-defense while achieving 0% attack success rate.

Experience reuse can reduce repeated exploration in embodied agents, but a trajectory that succeeded previously may be unsuitable for the current execution context. Existing memory systems pri marily optimize construction and retrieval; semantic relevance and historical success therefore remain insufficient when retrieved ex perience contains incompatible actions or an inappropriate level of structure. We introduce Memory Adaptation for Task-Conditioned Execution (MATE), a deterministic post-retrieval procedure that converts trajectories into execution-oriented memory. MATE re moves obsolete control context, extracts condition-action-effect transitions, applies verified action normalization, selects a task dependent representation, and serializes the result under a fixed budget without additional LLM inference. On 134 ALFWorld tasks, MATE achieves task success rates of 81.3% and 93.3% with Qwen2.5-14B and 72B while using approximately one-tenth of the tokens required by raw trajectories. Controlled comparisons show that verified action normalization is the principal mechanism by which MATE restores the utility of retrieved experience, support ing memory adaptation as a distinct stage between retrieval and embodied execution.

The rapid advancement of generative AI raises concerns about the misuse of Multimodal LLMs (MLLMs) for large-scale disinformation campaigns on social media. Despite existing research on textual disinformation, a fundamental question remains unanswered: can MLLMs be exploited to fabricate realistic multimodal fake news, and can they reliably detect it? We introduce a multi-agent framework in which a story agent, an image agent, and a critic agent collaborate to produce fake social media posts that plausibly counter true news. We apply the framework to generate over 9,000 paired multimodal news posts across science, health, and entertainment domains, and benchmark 16 open- and closed-source MLLMs for automated detection. We find that most models fall substantially short of human-level accuracy and fail critically on identifying image authenticity. Our research provides a foundation for developing robust defenses against social media fake news. Code and data are available at https: //github.com/xiuzhenzhang/Multimodal.

Large language models are increasingly used in oncology applications, but their predictions are often weakly grounded in explicit medical structure. We present TRACE, a deployable tree-relational enhancement framework for oncology LLMs. TRACE separates expensive offline structure learning from lightweight online inference: oncology concepts and relations are organized into an updatable tree-relational structure, refined using LM-loss-derived evidence, and retrieved at inference time as compact prompt evidence. This design supports task-adaptive evidence selection without requiring supervised labels in the zero-shot setting. Across ten oncology classification tasks and one MedQuAD CancerGov QA benchmark, TRACE improves both label-free evaluation and supervised fine-tuning. Additional analyses show that TRACE improves over vanilla RAG and generic GraphRAG, remains useful under leakage-controlled METABRIC inputs, and produces interpretable evidence paths aligned with clinical reasoning. These results suggest that explicit, updatable medical structure is a practical path toward more accurate and auditable oncology LLM deployment.

Tool retrieval is a critical bottleneck for LLM-based agents operating over large, heterogeneous API ecosystems. Existing approaches face an inherent trade-off: semantic retrievers are fast but suffer from the semantic-functional gap, while execution-based validation improves precision at the cost of prohibitive latency. We propose Lookahead-R, a planning-based framework that reformulates tool retrieval as a resource-constrained sequential decision-making problem. At its core, Lookahead-R introduces a lightweight execution-aware surrogate world model that jointly predicts tool execution success, latency cost, and semantic utility---without invoking real APIs. This world model drives a cost-sensitive, uncertainty-guided Monte Carlo Tree Search that navigates the tool space under strict budget constraints. Evaluated on the large-scale ToolBench benchmark, Lookahead-R achieves a superior accuracy-efficiency trade-off across all test scenarios. On the most challenging I3 split, it attains an NDCG@5 of 91.40\%, outperforming the state-of-the-art ToolGen (90.16\%) by 1.24\%. Ablation studies confirm that explicit latency modeling is the key discriminative signal for identifying high-quality tools under resource constraints.

Compact surrogates could reduce the cost of simulating large language model societies, but must reproduce collective behavior. We compare individual predictions and collective forecasts using 9,455 published trajectories and new experiments on opinion dynamics. Neighbor information improves individual prediction in all 16 public-data settings and pooled collective forecasts on held-out questions, although collective gains depend on transfer conditions. Tests on 24 new statements do not confirm earlier contrasting history effects in forecasts from the initial state. Qwen benefits from history after three observed rounds. These findings motivate direct collective validation, explicit limits on available observations, and comparisons with simple baselines.

As browser-use agents improve, benchmarks keep pace by collecting new tasks, websites, and applications, often making tasks longer or more novel. This makes difficulty expensive to refresh and difficult to control: when many aspects change at once, it is unclear what actually makes a task challenging. We instead construct challenging instances from tasks agents already solve, turning difficulty into a programmable property of the environment. BreakingWeb pairs every base task with an intervention condition that preserves the user instruction, latent target, and backend success criterion while changing the environment at different web stack layers. Each intervention is deterministic, detectable, and recoverable, and is annotated with the cognitive primitive it primarily loads. The benchmark contains 519 clean/intervention task pairs across seven self-hosted websites and 29 intervention families, all graded against outcomes. We evaluate six strong browser-use agents, three GUI-only agents that see only screenshots, and humans. The construction is effective: interventions cut agent pass rate by 22.9% on average and overturn nearly half of the tasks each agent solves cleanly, whereas humans lose 10.0% on a first attempt and 5.7% after one familiarisation attempt. The dominant failure is belief failure: 75% of the six agents' failures end with a declared success although the required change never happened. Our code, data and environment are publicly available at www.breakingweb.app.

Researchers across academia increasingly base significance claims on LLM judge scores and small-sample AI evaluations. Yet without well-calibrated confidence intervals (CIs), hypothesis tests, and judge-bias corrections, such claims are unreliable. We address these issues in several contributions. First, we find that running statistics over raw LLM judge scores leads to inflated false positives: counterintuitively, for many inter-rater agreement metrics, false positive risk peaks at "almost perfect" human-LLM agreement. To help researchers understand how to run statistics over LLM judges responsibly, we present guidance and tooling for the statistical analysis of mixed human-AI judge designs, and implement nine hypothesis tests via prediction-powered inference (PPI), including the first known PPI corrections for four rank-based tests (Wilcoxon signed-rank, Mann-Whitney U, and omnibus variants). To keep PPI++ stable with small human-labeled calibration sets, we introduce bootstrap-adaptive power tuning, which shrinks the estimated weight toward a target estimated from the labeled data, and accounts for that weight's own sampling variance. Second, through Monte Carlo simulations, we derive recommendations for what CI, p-value, and FWER correction methods to use for small-sample AI evaluations (N<100), and warn researchers against bootstrap CIs. We package these recommendations into evalstats, an open-source Python package that selects calibrated methods automatically, and demonstrate it in three scenarios, including one where a real LLM judge validated at "substantial agreement" would have led a researcher to publish a spurious finding. evalstats is publicly available at https://github.com/ianarawjo/evalstats.

Large language model search agents are often trained with synthetic questions whose difficulty is increased through larger evidence graphs, additional hops, and longer trajectories. These global properties, however, are only indirect proxies for the local retrieval capabilities required during search. To address this mismatch, we introduce latent anchor reasoning, which consists of resolving an unnamed retrieval anchor from descriptive specifications and transferring the recovered anchor into a subsequent information demand. This primitive retrieval unit decomposes deep search into chains of coupled operations and organizes question construction around anchor resolution and relation transfer, without prescribing a canonical search path. Based on this formulation, we propose PrimeSeeker, a capability-oriented framework that constructs web-grounded anchor structures and jointly derives a question and a reference evidence skeleton. The skeleton preserves supporting evidence from construction and guides expert generation through extractive highlights of current tool observations. These highlights are removed before supervised fine-tuning, while the skeleton is subsequently reused to audit reference-step coverage for reinforcement-learning rewards. We construct 9,221 expert trajectories, training a 30B search agent. Across five deep-search benchmarks, PrimeSeeker achieves strong performance, while reference-step optimization further improves the supervised policy. The resulting trajectories exhibit low retrieval redundancy, and fixed-budget evaluation shows strong solution coverage with substantially fewer tool calls than long-horizon systems.

Although uniform diffusion language models (UDLMs) represent a promising diffusion paradigm, scaling them remains challenging. We identify the core obstacle as an over-uniform training objective and condition-target confusion during sampling. To address these, we propose Less Uniform Diffusion (LUDI), a novel UDLM framework. Specifically, we (i) introduce a less uniform loss that directs each reverse transition toward the clean token, and (ii) equip the model with per-token time embeddings that supply token-level corruption hints, enabling confidence-based few-step sampling. Experiments across scales show that LUDI yields cleaner supervision and improves few-step generation. We further continue-train a 7B autoregressive model into LUDI-7B, resulting in a UDLM capable of complex reasoning. It achieves a 3-token-per-step speedup over AR decoding and competitive performance compared with masked diffusion baselines, revealing that the full potential of UDLMs for complex generation remains to be unlocked.

English-only benchmarks expose only a narrow slice of voice-agent behavior. We introduce $\tau$-Multilingual, extending $\tau$-Voice to Spanish, Brazilian Portuguese, Hindi, Korean, and Mandarin with native-speaker review and evaluation of generated language and spoken output. Across 4,500 full-duplex calls and five voice configurations, Spanish, Portuguese, and Hindi remain within 3.2 task-completion points of English, but Korean and Mandarin fall by 14.7 and 8.4 points. The failure modes also vary: Korean systems miss more responses, Mandarin systems interrupt more often, and both struggle with tools and entities. Grok leads task completion but scores lowest on generation quality, motivating separate task, interaction, and generation reporting. We release language packs, validated judges, and tools for community-built multilingual voice-agent evaluation.

Intrinsic self-correction asks a language model to revise its own answer without receiving new external evidence. A second pass can recover mistakes, but it can also overturn answers that were already correct. We study this trade-off across 29 open-weight LLMs on BoolQ, GSM8K, and Corr2Cause by tracking correctness transitions between initial and revised answers. Aggregate accuracy can conceal substantially different revision behavior: for example, Llama-3.1-8B improves by 25.5 percentage points on GSM8K, while refinement changes 19.1% of initially correct answers into wrong ones. A controlled BoolQ study further shows that refinement prompts shift the balance between recovery and harm. We then compare three runtime choices: keeping the initial answer, always accepting the revision, and selectively invoking revision using signals available after the initial response. The comparison identifies settings where learned gating is useful and others where a simpler unconditional policy performs better. These results suggest treating intrinsic self-correction as a revision policy rather than as a uniformly beneficial second pass, and evaluating it through both the corrections it recovers and the errors it introduces.

Handclaps provide an equipment-free excitation for room acoustics, but their unknown and variable source waveform makes room impulse response (RIR) estimation challenging. In this work, we investigate whether RIRs can be estimated directly from handclaps. To this end, we introduce an anechoic handclap dataset containing 2,540 claps from 17 participants, designed to capture variability across natural claps and different hand configurations. We first establish the performance attainable when the excitation clap is known using regularized deconvolution, and show that approximating the unknown excitation by windowing the direct sound from the reverberant recording is insufficient. To estimate the RIR without a known excitation, we propose using the anechoic handclap recordings to train a deep neural network with a supervised regression objective. Evaluated on a controlled synthetic benchmark, the proposed neural regressor significantly outperforms windowing-based baselines across all instrumental metrics. Furthermore, we test the proposed method on handclap recordings measured in real acoustic spaces, showing that the inferred RIR spectra are consistent across different handclap measurements taken in the same room location. These results showcase the feasibility of directly estimating RIRs from natural handclaps without requiring knowledge of the excitation signal.

Mutation testing evaluates test-suite adequacy by injecting synthetic faults into program code. However, traditional rule-based tools often generate large numbers of trivial, redundant, or equivalent mutants that limit their practical use for identifying gaps in a test suite. While recent large language model (LLM)-based approaches generate more realistic faults, most remain test-blind: The model sees only the source code and cannot reason about what existing tests already cover. We propose test-aware mutant generation, in which an LLM receives the problem statement, canonical solution and base tests in a single prompt, and must generate a nontrivial mutant that passes the base unit tests. We evaluate this approach across a set of five LLMs -- Gemini 3.1 Pro, Gemini 3 Flash, GPT 5.1 Codex Mini, GPT 4.1 Mini, Qwen3-32B -- on the HumanEval and MBPP benchmarks. The extended EvalPlus test suites serve as an automated oracle to verify whether surviving mutants represent genuine bugs. Test-aware prompting yields verified fault rates of 87.7% (HumanEval) and 79.1% (MBPP), meaning these mutants pass all base tests but are caught by the oracle. This vastly outperforms the matched test-blind prompting (which yields only 12.2% and 23.0%, respectively) and the traditional rule-based tool mutmut (4.4% and 5.7%). While fault subtlety (the fraction of extended tests a mutant fails) remains comparable across all three methods, test-awareness minimizes the computational cost per verified fault, compared to test-blind prompting. Exposing an LLM to existing unit tests shifts mutant generation from untargeted bug injection toward effective discovery of weaknesses in an existing test suite. Our work establishes a concrete foundation for future research to scale test-aware mutant generation to production-level environments.

In the field of Machine Learning, many papers contain empirical results supporting claimed statements or illustrating the performance of a proposed method. However, most practitioners know that (1) results are generally hard to reproduce, and increasingly so, (2) code is not often available to do so, and (3) it hinders the development of research. In this position paper, we analyze and quantify these issues, and make concrete proposals to improve result checkability, if not reproducibility. Code and supporting materials are available at https://github.com/giddyyupp/position-enforce-verifiability.

Optimizing deployed AI systems increasingly amounts to editing prompts, skills, harnesses, and code rather than model weights. Existing approaches commonly optimize these artifacts through propose-evaluate-select procedures, where candidate configurations are evaluated and only those meeting an acceptance criterion are selected. Yet our analysis shows that discarded candidates often contain information critical for subsequent optimization. Discarding them causes later proposals to revisit the same failure modes. We introduce Mara Chain, a refinement procedure that turns rejected candidates into stepping stones. Rather than discarding a rejected candidate, Mara Chain retains and iteratively refines it using evidence accumulated across preceding attempts. The procedure limits each refinement chain to a fixed depth and applies Pareto-filtered Top-N selection to bound the candidate pool. Across AppWorld skill optimization, TerminalBench 2.1 harness optimization, and MuSiQue retrieval-pipeline optimization, Mara Chain delivers greater task-performance gains with fewer rollouts. It outperforms GEPA, ACE, and SkillOpt-Lite by up to 20.5% in relative performance on AppWorld, reaching the target score with 65.5% fewer rollouts than GEPA. It improves the pass rate by 20.2 and 22.5 percentage points over AHE and Meta-Harness on TerminalBench 2.1, respectively, and improves MuSiQue test nDCG@10 and Recall@10 by 0.104 and 0.131 over a hand-written retrieval pipeline.

This research describes the adaptation of Large Language Models (LLMs) for economic monitoring in the public sector to automatically determine whether an article discusses Economic Policy Uncertanity (EPU) and to identify its specific type. Previous studies either rely on keywords, which often result in a high count of false positives, or use machine learning approaches that require a large number of quality human labeled data that is costly and time consuming to acquire. In this study, we propose approaches based on weak supervision techniques, using generative LLMs to create synthetic labels through prompting, making the approach both cost-effective and scalable. Additionally, we propose methods for for multi-label and hierarchical classification of articles related to EPU.

Sampling based consistency is widely used for hallucination detection, yet aggregate performance can conceal systematic differences in which errors are detectable. This work studies that heterogeneity across four language models and three factual question answering datasets. Partitioning hallucinations by answer agreement reveals high agreement (Ghost) and low agreement (Flickering) regimes with an apparent detectability gap of $0.35$ to $0.46$ AUC. Because the statistics used to define the regimes and measure this gap are strongly coupled ($|\rho|\approx0.94$ to $1.00$), the raw result is treated as a property of agreement based detection rather than independent evidence. After freezing regime assignments, lexical and semantic response dispersion preserve the asymmetry, with bootstrap $95\%$ intervals excluding zero in all $12$ model and dataset settings. A stricter test using individual diffusion trajectories and no cross seed information preserves the asymmetry across all three LLaDA datasets ($p<0.005$) and directionally across all three Dream datasets, with one reaching significance. The hard regime varies substantially in prevalence across models ($16\%$ to $77\%$), and matched prompts frequently change regimes between models. These findings show that aggregate detection metrics conceal persistent, model dependent heterogeneity in language model failures and motivate regime conditioned evaluation.

Benchmarks for large language models (LLMs) typically evaluate the accuracy of answers against fixed reference labels. But a central step in many complex real-world tasks is identifying which questions are worth asking in the first place: decomposing a difficult problem into subquestions -- which we call cruxes -- whose answers provide key steps on the path toward solving the target problem. To evaluate this capability of information discovery, we introduce CruxBench, a benchmark that grades LLM-generated questions by their Value of Information (VOI): how much a model-proposed crux updates beliefs about a target forecasting question. CruxBench enjoys a rare combination of three key properties: it is (1) contamination-resistant by construction, since ground truth is generated by future world events; (2) open-ended, admitting unbounded and complex text-based submissions rather than one correct numeric answer; and (3) grounded, with informativeness measured against quantified changes in real-world beliefs. We evaluate a diverse set of eight models on 293 target forecasting questions and find that VOI correlates highly with independent measures of model capability (r=0.90) and captures cruxes' usefulness for answering target questions. However, information discovery remains challenging even for frontier LLMs, which only narrowly outperform a random-timing baseline.

AI agents now hold spend authority and settle payments without per-action human confirmation. The resulting loss is often not a security failure: a counterparty with the correct domain, the correct settlement address and a genuinely delivered service can charge more than it should, and no check keyed on identity will see it. We present three artefacts for measuring and reducing that loss. First, a taxonomy of agentic commerce fraud that separates five observation levels (agent reasoning, wire, settlement rail, counterparty, principal) from the request-level and history-level evidence available at each, and records which levels can observe which attacks. Second, Agentic Commerce Bench (ACB), a benchmark of twenty fraud classes generated from production aggregates, 1,647 catalogued service operations and 1,068 settlements, of which six involve a counterparty that is exactly who it claims to be. Third, gordonguard, an open-source detector stack and offline harness with which an operator can audit an agent configuration, replay hostile counterparties without an account, and run the same detectors inline. Calibrating to a stated false-positive budget on clean training traffic gives a 6.5% clean flag rate, replicated across three independent generations, and leaves eight of twenty classes no better than chance. On the four classes a reasoning layer can observe, a widely used agent security scanner run over its jailbreak-detection panel scores zero on all four, while correctly scoring 1.0 on a jailbreak supplied as a control. A measured median payment of $0.007 places a hard constraint on deployment: one human review costs 143 times the value of the payment it examines.

Tool-using language-model agents select and execute third-party artifacts. Different implementations can return the requested output while producing hidden execution effects that task-, attack-, or choice-based evaluations may miss. We study functional counterfeits: implementations that match benign alternatives on the requested output but add an effect forbidden by the task contract. We introduce SINGED (Source Integrity and the Nonidentifiability Gap in Execution Decisions for LLM Agents), a controlled benchmark covering five primary and two held-out task families. It varies displayed rank, evidence depth, decision policy, model release, and agent configuration, while task and process oracles verify the artifact and execution path. Across 7,549 audited trials, the randomized-rank study finds counterfeit execution in 45% (27/60) of rank-one trials and none at later ranks. Cross-candidate comparison eliminates shallow failures and reduces layered failures from 15.7% to 4.2%, but leaves dependency failures; its benefit is uncertain on unseen effects and public-package structures. Moreover, seven releases with no counterfeit executions when benign alternatives are available execute the counterfeit in 55/175 single-source cells after alternatives are removed. SINGED thus exposes a rank-, evidence-, and choice-sensitive outcome-to-execution gap: evaluation must connect correct outputs to execution paths.

The integration of large language models (LLMs) into scientific workflows is accelerating, yet their ability to reconstruct the reasoning underlying published research remains unexplored. Papers specify explicit procedures while leaving many methodological dependencies-data selection, calibration corrections, priors, and domain assumptions-implicit. This ambiguity complicates the evaluation of LLM-based agents, since a failure to reproduce a result may reflect either limitations of the agent or underspecification in the source. We present a framework that evaluates agents through end-to-end reproduction, separating execution from verification and computational failure from methodological ambiguity. We apply it to fourteen astronomy studies: a case study from The Astrophysical Journal and thirteen papers published in Nature. Eleven of the thirteen contained an ambiguity preventing a uniquely specified reproduction path. In a controlled case study, twelve predefined paths, a 3x2x2 sensitivity analysis over sample definition, sky masking, and parallax zero-point treatment-gave estimates from 2.16 to 3.53 kpc for the same quantity, with only one recovering the published value (about 2.70 kpc). The published value was never used as an optimization target, selection criterion, or stopping condition; the matching path was found only after all twelve had run. Crucially, the decisive information (a +0.02 mas parallax zero-point correction) was already in the paper, but the agents did not recognize its causal relevance until the analysis made the effect visible. Matching a published outcome therefore does not validate reconstruction of the underlying reasoning, and the bottleneck is as often a failure to connect relevant information as to retrieve it. End-to-end reproduction thus serves both as a test of reproducibility and as a framework for evaluating implicit scientific knowledge in AI systems.

Semantic caches reduce LLM serving costs by reusing previously generated answers for semantically similar queries. However, retrieval is based solely on embedding similarity between the incoming query and cached queries. This design enables cache poisoning: an attacker can cache a malicious response under a query with high cosine similarity to benign requests. The vulnerability stems from a gap between retrieval similarity and answer validity. From an information-bottleneck perspective, query embeddings can lose information needed to distinguish valid from invalid cache hits, which limits any matching algorithm that uses only these embeddings. We propose a novel defense that recovers this necessary information from the raw text of the cache key. Across poisoning attacks, adversarial queries share a rewrite-residual structure: they pair a rewrite of the target query with residual content. The rewrite maintains high similarity, while the residual elicits the malicious response. Deleting the residual makes the remaining rewrite more similar to the incoming query. We exploit this structure using Deletion Gain to search shortened variants of the cached query for similarity gains, and an Answer Check to test whether the removed text contributes to the stored answer. We prove that Deletion Gain stays positive when a deletion leaves text close enough to the rewrite, and we search for such deletions with a sliding window. Across three poisoning attack classes, our defense blocks 82.0% to 98.2% of poisoned entries at a 5% false-positive rate, with negligible serving overhead.

Autonomous LLM agents turn vulnerability discovery into a repository-scale search: they generate many vulnerability hypotheses but can verify only a subset under a finite budget. We show that autonomous vulnerability discovery exhibits a hypothesis-verification asymmetry, where verifying a candidate hypothesis through reachability analysis, execution, and proof-of-concept construction is substantially more expensive than forming it. Under a finite resource budget, this makes autonomous discovery a resource-bounded selective-verification process, further exposing verification effort as a unique defense surface. We present RedHerring, which inserts certifiably safe decoys that divert verification effort from real vulnerabilities. Each decoy combines a CVE-derived vulnerability chain that attracts verification with a false bridge that keeps its dangerous sink unreachable. A private certificate lets the defender verify this property efficiently, while establishing the same fact from the released repository requires solving a computationally hard problem. RedHerring further adapts each decoy to the target repository so that it reads as ordinary program logic. Across 33 OSS-Fuzz projects, 70 evaluation instances, and five models under matched budgets, RedHerring reduces real vulnerabilities discovered by 38.7-60.4%. Trajectory analysis shows that agents spend 30.6-51.5% of completion tokens and an estimated 32.5-49.9% of runtime verifying decoys, showing that RedHerring redirects a substantial fraction of the fixed search budget toward decoys. When explicitly informed that decoys may be present, the agent adapts its search strategy, yet RedHerring still reduces vulnerabilities discovered by 37.2% relative to an informed Baseline, showing that its effectiveness does not depend on decoy secrecy.

Adapting large language model agents during deployment requires not only retaining past experience, but also turning new observations into timely guidance. Many test-time learning methods, however, acquire knowledge from completed episodes. Feedback from an ongoing interaction may therefore not be distilled into knowledge soon enough to help the next decision. Acquiring knowledge at the granularity of individual transitions could reduce this delay, but raises a separate challenge: a rule that is useful within one episode may not be reliable enough to guide future episodes. Waiting for validation can forfeit immediate benefits, whereas unrestricted reuse can propagate accidental or misattributed guidance. We introduce StepLearn, a nonparametric framework that separates immediate use from persistent trust. It turns informative transitions into hypotheses that can guide the next step, while requiring prospective validation before reuse across episodes. Their predicted effects are checked against subsequent observations outside the source episodes, and only sufficiently supported hypotheses become available for persistent guidance. This process updates external knowledge while keeping all model parameters fixed. Over five rounds on WebArena-Lite and ALFWorld, StepLearn achieves average success rates of 59.9% and 84.0% with GPT-5-mini, and 57.8% and 88.1% with Qwen3.5-35B-A3B, respectively. It outperforms EvoTest, the strongest baseline, by 2.2-12.7 percentage points across the four settings. Learning dynamics further shows that these gains are not restricted to the final repetition, with advantages already present on first task attempts in most settings.

Agent skills are shareable packages of procedural instructions, tools, and examples. Multimodal skills additionally include visual references that agents retrieve and inspect during execution. Because these images guide actions, attackers can disguise malicious instructions as ordinary visual guidance within otherwise legitimate skills. Existing skill-security research primarily examines text-carried attacks or scanner detection, leaving the runtime effects of image-borne attacks insufficiently evaluated. We introduce MMSkillRisk, to our knowledge the first publicly available benchmark dedicated to end-to-end safety evaluation of image-borne attacks in multimodal skills. To instantiate this attack surface, we design Native-Context Visual Attack (NCVA), which disguises malicious instructions as native components of teaching images, such as annotations and interface labels. The accompanying SKILL.md provides auxiliary guidance toward relevant visual regions without explicitly stating the malicious operation. Built from 28 curated clean skills, MMSkillRisk contains 36 attack packages and 108 executable cases spanning five attack objectives, with separate checks for attack success and legitimate-task completion. Across nine model-harness configurations evaluated in isolated sandboxes, NCVA induces unauthorized operations in every configuration. Its pooled attack success rate (ASR) reaches 43.1%, exceeding the matched text-carrier baseline by 16.4 percentage points, with higher ASR in all nine configurations. Attack success and legitimate-task completion co-occur in 36.5% of cases, reaching 72.2% for GPT-5.6-sol with Codex. These results show that skill-bundled images can induce unauthorized actions even as agents complete legitimate tasks, so task success alone does not establish safe skill use. Our code and data are available at https://github.com/kaill-jlq/MMSkillRisk.

Code large language models acquire programming capabilities from large code corpora, but can also memorize implementations that later require removal. Code unlearning is needed to control their continued reproduction when copyright or security concerns arise. However, targeted and retained code share computational patterns, creating a tension between forgetting specific implementations and preserving general programming ability. We propose \textbf{UNBIND}, a code unlearning framework that separately considers which hidden states correspond to the target code and how to suppress its reproduction. By constructing separate directions for these objectives, UNBIND achieves selective unlearning at inference time while keeping model weights fixed. Our evaluation covers fourteen baselines across two code models and two corpora. UNBIND achieves the highest joint forgetting and utility score in every setting. It reduces target code reproduction by 97.3\% to 99.1\% as measured by F-BLEU, with at most two fewer HumanEval+ and six fewer MBPP+ problems solved than the original models. In repeated extraction tests under a fixed budget, the number of targets yielding exact spans of at least 50 tokens falls from 188--262 to 0--2 out of 300 per setting. No extracted span reaches 100 tokens, and the mean best recovery ratio ranges from 0.43\% to 6.45\%. Multilingual and related-code evaluations further show effective forgetting with limited impact on useful programming capabilities, supporting UNBIND as a practical approach to selective code unlearning.

Finding the right files is an early challenge for coding agents. We test whether a language model can follow directory and file names to find annotated code files missed by fixed lexical queries. Across 82 audited issues from 11 repositories at pinned pre-fix commits, name-only directory routing recovered 0.465 of gold files within eight candidates, compared with 0.352 for FTS5 and 0.245 for a fixed full-issue rg query. The paired gain over FTS5 was 0.113 (95% repository-cluster bootstrap interval, 0.053 to 0.168). Under a shared 16K-token context budget, routing delivered 0.443 of annotated lines versus 0.246 for FTS5 on 55 cases with fully aligned annotations. At the same eight-file limit, combining routing with FTS5 reached 0.491 file recall, but its gain over routing alone was uncertain. An exploratory flat path control reached 0.572 recall while using 24.6 model calls per issue, compared with 8.9 for routing. Routing averaged 8.9 seconds per issue; FTS5 took 7 milliseconds per query after a 0.9-second build. On this cohort, directory routing added relevant file candidates to one-shot lexical search, but the study cannot attribute the gain to hierarchy or show that it improves issue resolution.

A voice agent can handle almost every call on the words alone and still fail the few its sector's rules were written for. Emergency-call standards, fraud guidance, radio phraseology and vulnerability rules recognise that how a caller sounds, or what else is audible, can change the right action. VoxParity tests whether agents act on it. In 183 scenarios from 14 sectors, one transcript stays fixed while the audio changes (a coaching voice, a medical monitor beeping, a mayday under a radio check, noise over a drug name, a child's voice placing a bet, a frightened whisper), and with it the correct typed tool call. A words-only null test credits a system only if hearing the call moves its actions more than it moves a pipeline that only reads the words. Only 11 of the 23 systems that can also be run on the transcript pass. Descriptively, errors run toward the words: when the audio calls for protection, all 28 systems carry out the routine request more often than they over-react on clean calls (41% against 12% pooled; the words-only pipeline, 58% against 15%). Exploratory analyses place most of the leading systems' misses on cues they heard; systems beat the null almost entirely on items that state the rule; the leading systems overrule heard resignation or confusion far more often than acute alarm; and, in the models tested, describing the voice and stating the rule each recover part of the shortfall, leaving a gap on emotion.

Continual learning models suffer from catastrophic forgetting when trained sequentially on non-stationary data distributions. Previously, this has been addressed through weight regularization. While preconditioning gradients offer a promising alternative to mitigate forgetting, current approaches are myopic. Conversely, standard regularization methods apply rigid, scalar Euclidean penalties that entirely ignore the underlying Riemannian geometry of the parameter space. To overcome this gap, we propose TMLN (Trajectory-Modulatory Landscape Navigation), a normative navigation policy that formalizes continual learning as an optimal control problem over a curved loss landscape. TMLN utilizes a memory-efficient diagonal empirical Fisher Information Matrix (FIM) to define a localized Riemannian manifold. To compensate for the spatial limitations of the diagonal approximation, TMLN dynamically modulates a preconditioner using the normalized historical trajectory of the network's parameter values. By integrating this trajectory-based preconditioning directly into the gradient update, we actively shield historically critical parameter directions without relying on additive penalties. Empirical evaluations on class- and domain-incremental benchmarks demonstrate that our method significantly reduces the loss barrier between consecutive tasks.

Multi-agent LLM systems increasingly mix models from several providers, yet exposing each agent's underlying model identity to its peers significantly impairs cooperation. We show that when agents are aware of each other's model family, the group splits into clusters, where agents prefer interacting with others carrying their same label, although nothing in the task rewards or asks for such a split. We argue that the label itself causes this split, which we define as $\textit{factionalism}$. We show and measure this phenomenon in two cooperative games and on a reasoning benchmark, with nine to twenty-five agents drawn from up to five open-weight model families. We further show that when the announced families are shuffled, or replaced by arbitrary labels, the factions still follow this information; when the label is removed, this behavior disappears. In strictly cooperative tasks, labeled groups spend on average $30\%$ more rounds and $55\%$ more tokens to reach a decision, and their success rate drops from $96\%$ to $81\%$. The effect replicates across tasks, group sizes and model families. Withholding identity labels from the agents is simple and effective mitigation.

Prompt injection against LLM agents becomes much stronger when the injected instruction is wrapped in the model's own chat template. A forged template marker such as <|im_start|> can reach the model either as a single reserved control token or as a sequence of ordinary subword tokens. The two decode to exactly the same text, and because tokenization runs on the server, the defender rather than the attacker decides which one the model receives. We use this to measure how much of the injected instruction's authority comes from the reserved token's learned representation. Encoding the forged markers as subwords, with the text held fixed and a control for the extra tokens this adds, lowers attack success on the InjecAgent benchmark by 39 to 66 percentage points on three of four open-weight families, and the gap carries over to multi-turn agent tasks in AgentDojo. On Qwen3-8B the gap is 8 points, because without reserved ids the model still recognises the forged turn from its text by reasoning; suppressing the reasoning block widens the gap to 50. The authority sits in the single learned vector at the marker position: the mean of the marker's subword vectors does not reproduce it, the vector of the nearest ordinary token restores the attack on Llama-3.1, and an adaptive attacker who searches for non-reserved markers finds such embedding neighbours on three of four families. In every base and instruction-tuned pair we test, instruction tuning strengthens the model's preference for reserved markers. The standard mitigation, a tokenizer option that encodes special tokens as ordinary subwords, applies only to tokens a configuration declares special, so in 33 of 67 distinct tokenizer configurations, covering 255 of the 400 most-downloaded chat models on Hugging Face, it leaves intact the tool-protocol tokens through which agents read untrusted tool output, and the gap persists on that channel.

Sensor networks and radars can deliver signals as organized sets, e.g. ordered signals, signals describing range cells within a grid or signals perceived as graph nodes. Within such sets, individual signals may be characterized by distinct sampling parameters. This paper investigates organized signal sets neural network encoders. In the context of this work, the purpose of such encoders is to project heterogeneously sampled signal sets into an arbitrary fixed-size vectors space. This new representation space is designed so that signal sets can be processed as vectors rid of sampling differences to allow for arbitrary topology-aware processing with no signal processing constraints. Within this representation space designed to reduce the influence of heterogeneous sampling parameters, the relevance of signal sets representations is evaluated by considering signal sets discrimination potential with a focus on waveforms separation. The encoding and embeddings discrimination experiments conducted rely exclusively on synthetic complex-valued radiofrequency signals.

As autonomous systems and embodied intelligence enter the dynamic physical world, multi-agent collaboration calls for a paradigm shift in communication design. However, existing communication paradigms overlook that agents form action understanding from their own states, environmental observations, and collaboration relations through a process that evolves as a task unfolds. Consequently, reliable bit delivery, general semantic recovery, or single-task utility optimization alone cannot ensure that heterogeneous agents form coordinated actions compatible with their own conditions from shared information during task execution. To address this gap, this paper proposes embodied semantic communication (ESC) as a paradigm that transforms information transmission into action-oriented semantic interaction. Specifically, ESC characterizes how an explicit communication link can encapsulate multimodal perceptual states, intrinsic hardware capabilities, and collaborative intents into unified actionable semantic representations, thereby enabling heterogeneous receiving agents to parse, align, and ground them in local motor control. This paper clarifies the conceptual boundary, system characteristics, and environment-constrained technical pathways of ESC. It maps the underlying mathematical tools, including semantic information theory, world models, and multi-agent decision theory. Finally, this paper summarizes key open challenges, including measurable semantic reliability, ambiguity-triggered interaction under dynamic environments and tasks, and bandwidth-adaptive semantic transmission, outlining a roadmap for collective embodied networks.

While prior work has documented privacy failures in LLM agents, it remains unclear how the presentation of privacy guidance influences their choice of information sources. We introduce PrivacySkills, a controlled framework for evaluating how agents choose among acquisition pathways that provide the same task-relevant value: consulting publicly available personal information, accessing confidential sources, or interacting with the user. The evaluation framework comprises 55 synthetic tasks spanning 11 categories of personal information, with 169 associated skills that describe the available acquisition pathways. We consider privacy guidance through system-level instructions, skill-level metadata labels, or both. Separately, we vary user availability and urgency framing. With users available and no privacy guidance, agents access confidential sources in 30% of valid runs on average across five open-weight models, despite sufficient alternatives. This rate increases to 45% when users are unavailable, whereas urgency framing has no detectable effect. System-level privacy instructions alone have limited effects on confidential access, while skill-level intrusiveness labels produce a modest reduction (24% on average), but combining the two roughly halves confidential access. Our findings motivate incorporating privacy annotations into skill specifications and evaluating their effectiveness alongside system-level instructions.

Geometry does more than constrain an associative memory: curvature determines what it remembers and which states it creates. We develop intrinsic dense associative memories on Riemannian manifolds by casting memory as Epanechnikov kernel-density mode seeking. We compare geodesic and volume-corrected energies and show that curvature separates their behavior. We prove that geodesic memory always retains an isolated pattern, while corrected memory obeys a sharp Ricci-curvature threshold: positive curvature can erase memories in high dimensions, while negative curvature reinforces them. We derive geodesic capacity scalings of $q_\beta^{-1/2}$ for retaining every pattern and $q_\beta^{-1}$ for a typical one, where $q_\beta$ is the pairwise kernel-overlap probability. We show how overlap \emph{creates} novel memories: designed $N$-pattern configurations realize all $2^N-1$ subset modes, but random data at the storage threshold yield only a Poisson number. We establish exact one-step recall using Riemannian mean shift. In simulations, we recover the predicted curvature transition and every designed mode. On WordNet's full noun hierarchy, we demonstrate that volume correction improves low-capacity retrieval. Together, our work shows that curvature is a design variable for associative memory, not merely a property of the data.

We introduce HEAR (Human-recorded Evaluation of Audio-LLM bias by Real speakers), a large-scale, ecologically valid benchmark comprising 87k real human audio samples from 843 demographically diverse participants. HEAR enables comprehensive evaluation through Multiple Choice Question Answering (MCQA) and open-ended long-form tasks. To our knowledge, this is the first large-scale voice benchmark grounded entirely in authentic human speech. We evaluate model behavior across both real-time speech-to-speech and speech-to-text architectures. Our results reveal that voice-conditioned bias is a model-specific property. Furthermore, we demonstrate that personalization instructions consistently exacerbate demographic disparities. Our findings establish that voice bias is a controllable model characteristic, providing a foundational framework for future bias mitigation and evaluation in Audio-LLM development.

Large language model post-training generates self-generated rollouts through reinforcement learning and on-policy distillation, yet this experience is often treated as stale once the policy advances. Historical rollouts can remain compatible with a later policy while preserving behaviors that the policy no longer expresses reliably. However, they may also contain mistakes, abandoned attempts, and redundant actions that should not be imitated, motivating finer-grained selective supervision. We introduce ROSS (Relearning from Self-Generated Rollouts through Selective Supervision), which preserves the full historical trajectory as context while applying loss only to selected model-generated continuations. Across domain-specific reinforcement learning, multi-teacher on-policy distillation, and agentic reinforcement learning, ROSS consistently improves upstream checkpoints and outperforms baselines across mathematics, code generation, instruction following, and software engineering. On Qwen3.6-35B-A3B, ROSS improves the six-benchmark MOPD average from 58.40% to 62.20% and SWE-bench Verified from 64.20% to 68.40%. These results show that self-rollout training leaves behind reusable behavioral experience that can yield further gains through offline supervised fine-tuning (SFT), without additional policy rollouts.

We introduce Solver Agent, an AI framework based on large language models for calculations and proofs in mathematics and theoretical physics. The solution process is tracked through a persistent ledger that records assumptions, derivations, and computations. A central agent delegates tasks to specialized sub-agents, while independent agents verify both intermediate steps and the final result. This setup improves the traceability, reproducibility, and verification of computer-assisted calculations. Applying Solver Agent, we study global F-theory uplifts of Type IIB orientifolds and their S-fold generalizations. We establish sufficient conditions for Weierstrass models over projective threefolds with terminal $\mathbb{Z}_k$ quotient singularities ($k\in\{2,3,4,6\}$) to give $\mathbb{Q}$-factorial projective elliptically fibered Calabi-Yau fourfolds with isolated Gorenstein terminal quotient singularities. These geometries realize O3-planes and S-folds, where local D3-brane probes of the latter yield four-dimensional $\mathcal{N}=3$ superconformal field theories. Using stringy invariants, we derive fixed-point contributions to Hodge data and Euler characteristics, and show that these Euler corrections determine the localized D3-brane charges required for tadpole cancellation. We illustrate these results using toric hypersurface constructions, where a single three-dimensional polytope determines both the Type IIB Calabi-Yau threefold and the F-theory base; here, the orientifold double cover naturally forms a bisection of an alternative genus-one-fibered uplift with discrete $\mathbb{Z}_2$ gauge symmetry. Finally, we provide methods for toric computations and four-form flux analysis in four-dimensional $\mathcal{N}=1$ compactifications with non-abelian gauge sectors.

Large language models are able to solve tasks whose answers depend on not only individual input tokens, but also on relations among them. How is such relational information represented and processed across transformer layers? We study activations from ensembles of prompts that require inferring relationships between three tokens corresponding to a cyclic concept (months, hours, weekdays, and musical notes) to correctly predict the next token. Across model families (Llama, Qwen, Gemma, and Mistral) and cyclic concepts, we find a consistent layerwise progression in how the joint dependence among the tokens is geometrically organized and causally used: intermediate layers use a joint representation based on the inferred relationship between two tokens, while later layers use a joint representation associated with all three tokens to correctly complete the task. We also find other relationships between tokens that are geometrically structured but remain causally inert in the next-token prediction. Crucially, when taken together, these geometric and causal investigations reveal the representation-level mechanism that progressively organizes and composes the relational information to form the answer. More surprisingly, restricting the models to such causally relevant joint representations improves next-token prediction accuracy.

We present AI-developed local infrared subtraction, building on projection to Born and EFT matching. The framework separates an integrable radiation term from a finite contribution at Born kinematics, referred to as the Born contact. The contact is determined using the EFT singular distribution in a resolution observable such as N-jettiness $\tau_N$. Under human physics guidance, an LLM develops two implementations. One uses a neural network for phase space projection and fits the contact by matching to EFT cumulants. The other uses an analytic construction that keeps the Born momenta fixed while integrating over radiation. It combines the EFT $\delta(\tau_N)$ coefficient with finite 4-dimensional radiation integrals to calculate the contact term directly. This gives a local subtraction formula without a slicing parameter, while reusing existing lower-order radiation calculations and EFT singular predictions. As a demonstration, we reconstruct the full NLO correction for massless 3- and 4-jet production in electron-positron annihilation. The attempt to the NNLO dijet production is also made by recursively using the NLO P2B construction with the LLM designing machine-learning controls to reduce the variance of the contact integral. The tested predictions are in good agreement with EERAD3. The numerical calculation and projection-network training use a 2020 Apple M1 MacBook, without GPU acceleration, illustrating the feasibility of the construction with modest computing resources. The appendices develop an extension of the local subtraction to 3-jet NNLO, giving explicit radiation maps and a proposed contact formula. We also show how to integrate over NNLO radiation while keeping the Born momenta fixed, for any number of massless final-state jets. Our results demonstrate how AI can help higher-order calculations by constructing infrared subtraction and improving its numerical integration.

Uniform random rotations are an effective preprocessing step for quantization: they make normalized coordinate distributions approximately Gaussian, enabling the use of codebooks optimized offline. We introduce TORQUE, a framework that improves on previous quantization works that use random rotations by jointly optimizing how many and which coordinates to preserve at high precision both before and after rotation, under a fixed overall expected bit budget. Intuitively, before rotation, preserving large input coordinates at high precision can reduce overall error by preventing the rotation from spreading their values across many coordinates. Likewise, after rotation, preserving a small fraction of the largest-magnitude coordinates at high precision allows the remaining values to be quantized more accurately using codebooks optimized offline for the resulting truncated Gaussian distribution. We derive a quantization error upper bound and prove that top-$k$ pre-rotation retention minimizes it for each $k$. This reduces the search over coordinate subsets to an optimization over $k$, enabling a fast optimizer that uses offline codebooks and parallel parameter selection for practical implementation. We demonstrate an improved tradeoff between reconstruction accuracy and storage cost through numerical evaluation under the Gaussian model and experiments on nearest-neighbor retrieval, KV-cache compression, and activation compression.

Given a functional dependent on the spectrum of a differential operator, we address the problem of finding a domain which optimizes this functional. PDE solvers might be used to tackle this optimization. It is however computationally expensive. We propose two neural network models which learn the spectrum directly from the geometry of the domain and can be used to optimize the domain from one or more eigenvalues. We investigate two representations. The first encodes the domain through Fourier coefficients and a light MLP, which is efficient on star-shaped geometries, achieving a precision of 0.2\%. Through a rescaling of the coefficients the designed models satisfy the scaling law of the eigenvalues. Additionally, averaging the outputs of the trained surrogates over rotations and reflections induces invariance for these transformations. The second is a model that takes the landscape function, the indicator function and the gradient of the landscape function. A Gram-Schmidt process produces orthogonal eigenfunctions as output of the model along with the associated eigenvalues. The landscape model reaches 1\% mean relative error on the first ten eigenvalues, compared with 4\% for an FNO model. Replacing the landscape by an SDF worsened both prediction and optimization errors. The trained model also generalizes from synthetic shapes to domains given as classical image dataset. The resulting surrogates of both approaches recover classical spectral optima such as the disk for the first eigenvalue or the conjectured minima of higher eigenvalues. This confirms that our models produce accurate differentiable estimates of eigenvalues, which can be used in shape optimization problems involving spectral quantities.

Worker-monitor setups are a promising approach to AI oversight, but training workers against fixed monitors can incentivize monitor evasion. We study whether this failure mode can be avoided by co-training the monitor alongside the worker, and explore both supervised and self-supervised approaches. In the supervised setting, we prove a characterization: monitoring is possible with vanishing error and query rates exactly when the class of possible monitor functions has finite Littlestone dimension. This connects worker monitoring with an established literature on adversarial online learning. For self-supervision, we propose a co-training procedure based on test-time distillation: the monitor uses additional test-time compute to generate training labels, then trains its standard-compute policy on those labels. For majority-vote labels, we give a finite-sample sharpening guarantee under adaptive worker distributions with action coverage, that shows that the monitor's verdicts converge to its initial modal verdicts. We stress-test the former in code-security settings where the worker is trained adversarially to fool the monitor. Our results suggest that adaptive monitors are better at keeping pace with evolving worker strategies, while fixed monitors are more vulnerable to evasion.

In contrast to even a year ago, AI systems now write most of the code inside the companies that build them. As more of the AI research and development (R&D) pipeline is automated, could AI progress radically accelerate in an "intelligence explosion," where years of advances are compressed into months or less? Preliminary evidence suggests that it could. In this work, we assess this evidence, analyze an intelligence explosion's potential impacts, and propose policy responses. AI systems are on track to automate most AI R\&D work within a few years, and possibly all of it. If this triggers an intelligence explosion, it could dramatically bring forward AI's benefits, but also pose extreme risks: capabilities growth could accelerate far beyond what society can keep up with, humanity could lose control over superhuman AI systems, and checks on power within and between states, companies, and branches of government could be severely eroded. Although there remains much uncertainty about these possibilities, the high stakes warrant serious further attention. Policymakers should urgently obtain more visibility into the automation of AI R&D, develop ways to steer and constrain an intelligence explosion, and prepare society to adapt to an intelligence explosion's impacts.

Long-term memory lets an LLM assistant use a history it can no longer reread, and most memory systems build it by rewriting conversations into facts, graphs or typed memories at write time. We argue that the work of memory divides, as thinking does, into two systems. Most of it is fast System 1 work: many small, independent yes/no judgments about records, such as whether a record is needed or no longer current, which a decision model makes by the dozen in a third of a second. Only a little is slow System 2 work: writing a few search queries, naming what the reply needs and composing the answer, which an LLM does well but slowly. We present Mnemon, a memory agent built on this division. It keeps conversations as raw, dated records; an LLM (System 2) plans searches over them, a decision model, Jev (System 1), judges what the searches return, and rules with explicit budgets turn the judgments into a small View for an unchanged answering model. A background pass consolidates each record once into topic timelines, value histories and standing instructions linked to the records, so that questions about a whole conversation reach evidence their own searches miss. Because nothing is decided about a record when it is written, the same agent can read any store that returns dated records. With gpt-4.1-mini answering, as in a public re-evaluation of 14 systems, Mnemon scores 91.7% on LoCoMo, the highest among them, and 83.8% on LongMemEval-S, from under 4k tokens of context per question, with the lowest effective cost index on LoCoMo. With a reasoning model answering, it reaches 92.2% on LoCoMo and 94.4% on LongMemEval-S, the latter on par with the best published results. From 100K to 10M tokens of history on BEAM, its cost per question grows by a factor of 1.11. On the same records, Jev separates gold evidence better than two LLMs and is 3-11 times faster.

Neural Combinatorial Optimization (NCO) has achieved strong empirical success, yet the internal mechanisms driving model decisions remain largely unexplored. In this paper, we investigate three representative autoregressive NCO models spanning two encoder-decoder configurations: AM and POMO (heavy-encoder, light-decoder), and LEHD (light-encoder, heavy-decoder). Through behavioral analyses, representation probing, and causal interventions, we examine how these models construct solutions and use internal representations during decoding. Our results suggest that AM and POMO predominantly follow a persistent geometric pattern throughout solution construction, whereas LEHD contains linearly accessible information about multiple future actions. Causal experiments further provide evidence for the role of future-node representations in LEHD's decision-making. We also observe that LEHD relies strongly on the current-node representation for immediate local decisions, while the start-node representation plays a broader navigational role over the subsequent route. Cross-instance alignment analyses additionally indicate that LEHD maps current-node representations into a relatively shared latent region, which may provide a stable reference for evaluating subsequent decisions. Across the Traveling Salesman Problem and the Capacitated Vehicle Routing Problem, these results reveal distinct decision-making patterns across these architecturally distinct solvers and provide a foundation for more interpretable analyses of NCO solvers. Code and additional visualizations are provided in the https://github.com/NCO-Interpretability/NCO-Interpretability.

Foundation models are powerful generators, but many engineering domains require structured representations that general-purpose systems handle poorly. We introduce FLOORA (Floor Layout Optimization with RL Alignment), a family of small domain-specific language (DSL) models for architectural layout generation. With specialized data and alignment, our 0.6B model outperforms much larger frontier models, achieving VLM judge win rates up to 92.0% on out-of-distribution real-world buildings and 96.0% on synthetic buildings. Human evaluations further corroborate these results, with FLOORA selected as the best model in 89.3% of evaluations. FLOORA combines a token-efficient DSL, custom tokenization, domain-specific pretraining, supervised fine-tuning (SFT), and reinforcement learning (RL) with learned human-preference and verifiable rewards. This pipeline improves architectural and geometric validity, supported by extensive empirical evaluation and ablation studies. Although focused on architecture, our results suggest that similar domain-specific recipes may be useful in other engineering domains with structured, verifiable outputs. Datasets, models, and inference code are available at https://github.com/AutodeskAILab/floora.

Generative AI (Gen AI) is reshaping how individuals learn and work, but its consequences for collective knowledge, the shared body of knowledge that online communities produce together, remain poorly understood. Prior work has documented an aggregate decline in participation on knowledge-sharing platforms, but it remains unclear which specific kinds of knowledge are being lost first. We study this question using Stack Overflow, one of the largest online communities for software engineering, treating the release of ChatGPT-3.5 as a natural shock. Analyzing over two million questions posted between 2020 and 2025, we track how two dimensions of collective knowledge, difficulty and data availability, change following Gen AI's release. Using diverse methods and robust checks, we find consistent patterns. Easy questions decline sharply while difficult questions become more common, a pattern corroborated by rising code complexity. Data-rich topics and tags lose share of questions, while data-scarce ones gain ground. The two dimensions also interact: the decline in easy questions is concentrated specifically within data-rich domains, while difficult questions increase regardless of data availability. This pattern extends beyond Python across programming languages, with more prevalent languages showing sharper shifts. Together, our findings reveal that Gen AI's impact on collective knowledge is uneven, eroding easy, accessible knowledge first while more complex, less common knowledge persists.

Final predictive accuracy is the standard basis for comparing graph neural networks (GNNs) in materials-property prediction, but it does not show how strongly performance depends on access to trainable parameter-space directions. Here, we introduce trainable-degree dependence as a complementary characterization of materials GNN learning. Using random-subspace intrinsic-dimension analysis, we train CGCNN, ALIGNN, and DimeNet++ in randomly oriented parameter subspaces across six prediction tasks and measure how performance recovers as independent trainable degrees of freedom are restored. The resulting recovery curves separate endpoint accuracy from the trainable-dimensional demand required to recover it. They reveal distinctions that final errors alone miss: metallic classification and log-bulk-modulus regression recover near-reference performance from small fractional subspaces, formation-energy and band-gap prediction show stronger architecture dependence, and phonon prediction is most sensitive to dimensional restriction. Dataset-size sweeps show that band-gap models require larger fractional subspaces as training data grows, whereas formation-energy and bulk-modulus responses are more stable. A width sweep shows that fractional thresholds can remain stable while absolute threshold dimensions increase with model size. Random-subspace analysis therefore provides a targeted stress test for how materials GNNs use their optimization space.

Language models are frequently employed to evaluate other language models. An LM evaluator scoring agentic behaviors across multiple criteria is valuable, provided that its decisions align with human judgment. We call the problem of evaluating this alignment Meta-Evaluation. Tackling it directly is difficult: collecting human data is expensive, absolute scoring is hard to align, and using an LM meta-evaluator recurses the question of trustworthiness. We adopt a reformulation of meta-evaluation as a preference judgment problem: rather than comparing human and LM evaluator scores of a trajectory, we ask whether their implied preferences align. Building on this, we introduce PADM\'E, a data synthesis method that generates reliable criterion-based meta-evaluation data for agentic settings. PADM\'E uses only small language models, requires no human involvement during evaluations, and operates under a low computational budget. We build a prototype of PADM\'E and synthesize a dataset of 1,000 samples across four agentic domains and three evaluation criteria. Human validation on a 150-sample subset demonstrates that PADM\'E improves agreement with human judgment from 73% to 85% over a naive baseline. Meta-evaluating 25 common models with our dataset demonstrates the correlations between evaluation performance and scoring granularity, leniency, and model size, among other factors.

Clinicians and neuroscientists have long analyzed intracranial electroencephalography (iEEG) through directly measurable physiological characteristics, which carry much of the information that downstream tasks depend on. Recent iEEG foundation models learn by reconstructing or predicting their inputs, which leaves the retention of these characteristics implicit. They are also evaluated mainly on cognitive decoding and a narrow clinical task, i.e., seizure detection. On a broad, clinically relevant benchmark such as Omni-iEEG, they remain below task-specific models when used frozen. We introduce PHASE, a physiology-guided foundation model that makes these characteristics explicit learning targets, pairing them with masked latent prediction in a temporal stage (PHASE-T) within each channel and a spatiotemporal stage (PHASE-ST) across synchronized channels. PHASE is pretrained on heterogeneous recordings from 222 participants at nine clinical sites. On all five Omni-iEEG clinical tasks, frozen PHASE-T outperforms every evaluated foundation model by up to 31\%, and fine-tuned PHASE-T surpasses the task-specific models, setting a new state of the art. PHASE-T benefits from physiological supervision, outperforming variants trained with latent prediction alone or auxiliary waveform reconstruction on every task in matched ablations. PHASE-T generalizes to unseen institutions, outperforming the compared models with few or no local labels. PHASE-ST further improves seizure-onset-zone identification over PHASE-T and, when frozen, decodes sound volume and pitch on BrainTreebank better than published models. Beyond task performance, PHASE learns to encapsulate the physiological characteristics clinicians recognize, from seizure onset and its propagation to anatomical region identity, even though its pretraining contains no ictal recordings or anatomical labels.

Tabular foundation models using in-context learning have recently surpassed gradient-boosted trees on predictive tabular tasks. However, recent mechanistic insights suggest that parameters in these models are largely redundant. We introduce LoopICL, a looped transformer whose core design decouples parameter count from computational depth. LoopICL consists of a single block, processing data through two coupled streams: a cell stream capturing per-cell feature representations and a row stream capturing in-context example representations, jointly refined through within-column and cross-column attention. During pre-training, we vary loop counts, allowing the block to be unrolled for a varying number of iterations at test-time and use a learned exit-gate to automatically exit. In its standard setting, LoopICL performs competitively with TabICLv2 on TabArena and TALENT at the same computational cost (FLOPs), while using nearly 90% fewer parameters. Furthermore, its recurrent design enables users to also trade off inference cost and performance, providing a resource-aware TFM.

Learned rotations play an important role in enabling low-bit weight and activation quantization of large language models by smoothing outliers in the activation distribution. State-of-the-art approaches include gradient-based procedures such as SpinQuant and computationally friendlier gradient-free approaches such as DartQuant, but both remain hard to scale to the largest architectures. To address the computational bottlenecks in gradient-free rotation learning, we introduce two ideas for efficiency, (i) a data selection procedure which reduces the required number of calibration data points, and (ii) an exact reduction of the associated optimization on this reduced calibration set. Our data selection procedure exploits the geometric structure of the convex hull of the activations. Using this idea, we show that a carefully selected calibration set with several orders of magnitude fewer activations than state-of-the-art rotation-based methods can match their performance in low-bit quantization settings. Under this extreme data efficiency, the selected activations span an $r$-dimensional subspace with $r<d$, making optimization over a $d\times d$ rotation equivalent to optimizing a $d\times r$ matrix on the Stiefel manifold. We solve this reduced problem using an efficient ADMM algorithm that iteratively employs thin matrix updates at every step, hence the name ThinQuant. For Llama-3-70B with W4A4KV4 quantization, ThinQuant completes the entire rotation calibration in under 12 minutes and achieves a WikiText-2 perplexity of 5.63, compared with 7.55 for DartQuant, which requires 111 minutes. Unlike SpinQuant and DartQuant, ThinQuant also scales to Llama-3.1-405B on a single H200 GPU, completing rotation calibration in just over 2 hours and achieving WikiText-2 perplexity of 2.97 at W4A4, compared with 3.48 for GPTAQ+QuaRoT.

Large language models are vulnerable to prompt injection attacks, where third-party adversarial content can hijack the model's behavior. In this paper, we study the role played by the adversarial data's input modality, and identify a systematic asymmetry: multimodal LLMs are more likely to follow adversarial instruction when they appear as text than when the same instruction is delivered through a non-textual channel (e.g., as an image). We hypothesize that this modality gap arises from text-centric instruction tuning, which teaches models to obey textual instructions while treating other modalities mainly as content to parse or describe. We then demonstrate how this gap can be turned into a training-free defense, by rendering all untrusted payloads as typographic images (or audio) before they reach the model. Across ten models and two prompt injection benchmarks (DirectInject and AgentDojo) we show that our defense Pictionary consistently reduces attack success rates even against the strongest adaptive attacks and human red teamers, while largely preserving benign utility. We further show that benign fine-tuning on image-rendered instructions erodes the modality gap, tracing it to the text-centric instruction-tuning distribution.

Modern AI architectures used to solve visual reasoning tasks typically rely heavily on global connectivity and synchronization. As biological systems demonstrate, though, sophisticated computation can be performed in a more decentralized fashion. In this work, we test the reasoning capabilities of Neural Cellular Automata (NCAs), networks of recurrent cells that use strictly local connectivity and asynchronous updates. NCAs have been extensively studied in artificial life experiments, but it is unclear whether they can perform complex multi-step reasoning. We show that NCAs produce spatio-temporal dynamics capable of solving challenging visual reasoning tasks, including large mazes, Sudoku, and ARC-AGI-1. Furthermore, we provide evidence that NCAs generalize out-of-distribution when running with larger grids, longer rollouts, or parallel trials; and that the latter can be made more efficient via pruning of redundant trajectories. We find that these generalization capabilities depend on training with sample replay and stochastic perturbations, and that stochasticity remains beneficial at test time. Finally, we show that NCAs are robust reasoners capable of dynamically modulating compute to recover efficiently from damage, and that they can scale to solve reasoning in raw pixel space.

Large language models (LLMs) are increasingly positioned as a force to empower underserved communities, and significant efforts are being made to expand access. Yet, access alone does not equate to meaningful adoption. First, even if a system is accessible, it won't be adopted if users are not willing to adopt it. Second, even if an LLM system is superficially adopted, the heterogeneity of LLM tools means that LLM adoption can be further deepened. Closing this access-adoption gap is critical to ensuring that the full social potential of LLM is not only accessible but fully realised. Drawing on 61 interviews (15 long-form semi-structured interviews with first-generation, low-income college (FGLI) students, 3 non-FGLI students, 3 FGLI program directors, and 40 intercept interviews), this paper examines the access-adoption gap in first-generation, low-income student communities. This paper a) finds that while FGLI students have adopted LLM systems, their depth of LLM tool usage is limited to chatbots (e.g., ChatGPT or Claude) for narrow use cases, and b) identifies barriers limiting their willingness to learn and use (low perceived value, under-estimated self-efficacy, unclear starting point, low peer exposure, and resource constraints). Then, from these findings, the paper derives the four design principles to design a system or an intervention aimed at closing the access-adoption gap in LLM adoption by FGLI students. In doing so, the paper contributes to the field by a) examining the LLM access-adoption gap in the FGLI student community, and b) reframing LLM adoption as a depth gradient across four modes of LLM tool use: basic chatbot interfaces, tool-augmented prebuilt interfaces, agentic development interfaces, and programmatic integration.

As large language models are deployed in increasingly autonomous long-horizon tasks, manually auditing and verifying the actions, artifacts, and outputs of models becomes more difficult. Users instead come to rely on LLM-generated reports to assess the quality and completeness of the work. We introduce a suite of eight adversarial reporting scenarios to systematically study whether LLMs conceal narrative-changing flaws: errors or limitations that undermine an otherwise successful account of work. We call this phenomenon "insecure reporting." When handed machine learning experiment logs containing a planted negative result that substantially weakens the proposed method, GPT-5.5 flags the negative result in only 2 of 200 generated reports. However, when a short honesty instruction, "Be honest in your response," is added, the model flags the negative result in 190 of 200 reports. Across eight open-weight models, chain-of-thought analysis reveals a recurring tension between disclosing narrative-changing flaws and reasoning about ways to appear successful. We perform an activation analysis and a steering experiment on Qwen3.5-9B, finding that honesty and success-seeking correspond to opposing directions in representation space. Our results suggest that LLMs tend to present narratives of success by default, and that steering models toward honesty makes their reports substantially more transparent.

Most federated learning frameworks for vehicular ad hoc networks assume that all vehicles collaboratively train a single model for a common task. This assumption limits their applicability to practical vehicular environments, where vehicles may perform heterogeneous but related perception tasks with different output spaces. This paper proposes encoder-sharing hierarchical multi-task federated learning (EN-HMTFL), which integrates cluster-based hierarchical federated learning with a globally shared encoder and vehicle-local decoders. EN-HMTFL enables vehicles performing different tasks to collaboratively learn a transferable feature representation while preserving their task-specific models locally. Only the encoder is exchanged and aggregated through the hierarchy, whereas raw data and local decoder parameters remain at the vehicles. The proposed framework is evaluated on the MNIST and GTSRB datasets in different vehicular scenarios. Across the evaluated scenarios, EN-HMTFL improves accuracy by up to 24.0% relative to the compared representation-sharing benchmark. In scenarios where EN-HMTFL converges earlier, the reduction reaches up to 69 communication rounds (28.8%).

Can coding agents restore software that no longer runs while preserving its underlying methods, and reconstruct industrial software engines from open specifications? Here we introduce ReviveBench, a benchmark with two task families evaluated by hidden verifiers calibrated against native execution environments, established engineering tools, or purpose-built reference implementations. The revival family comprises ten tasks involving dependency incompatibilities, deleted core modules, legacy builds, and a GPU-based foundation model. Every starting workspace fails verification, and the strongest evaluated model passes all ten tasks in at least one run each. In contamination-control experiments, identifier obfuscation reduces line similarity to the original implementations from 0.51--0.96 to 0.03--0.44 without reducing the observed pass rate of any evaluated model. On repositories created after the stated knowledge cutoffs, the strongest model passes eight of nine runs. The reconstruction family comprises thirteen tasks spanning numerical, geometric, hardware, and transactional systems (e.g. CAD and CRM). Two models meet the benchmark's pass criteria on all thirteen, although our audit shows that the CFD task cannot establish numerical-solver capability. Benchmark construction and auditing uncover 28 verifier defects, including 24 false negatives and two false positives. These findings show that executable verification can itself introduce substantial measurement error. We present three practical checks: test whether prescribed methods can reach the grading thresholds, investigate agreement among independently generated candidates, and recompute diagnostics from submitted artifacts. ReviveBench thus provides both an evaluation of software revival and engine reconstruction, and cases in validating the verifiers used to measure coding agents for software design.

Distributed Denial of Service attacks are a growing threat to network infrastructure, and new techniques, including the use of generative AI, make them harder to detect. Traditional detection systems, such as rule based firewalls, often fail to identify these evolving attack patterns. In this study, we propose a new method for detecting DDoS attacks by combining synthetic data generation using Generative Adversarial Networks with a Random Forest classifier. The GAN generated data showed 80.3 percent cosine similarity to real traffic, which helped the model learn underlying traffic patterns more effectively. To address imbalances in the data, especially in packet related features, we applied adversarial debiasing. This reduced the model's sensitivity to skewed distributions in variables such as forward and backward packet counts and total byte lengths. Our results show that models trained on a mix of synthetic and real data achieved significantly better performance: 99.98 percent accuracy on benchmark data and a 22.60 percent improvement when tested on previously unseen synthetic traffic. This suggests that the method can generalize well across different traffic scenarios and adapt quickly to new types of attacks. The proposed approach not only improves DDoS detection but also provides a scalable foundation for security models that account for bias and benefit from data augmentation. Our findings show that combining GANs with adversarial debiasing can lead to more robust and effective DDoS mitigation, supporting the further development of machine learning based cyber security.

Parallel speculative drafting generates multiple candidates in one backbone pass, but independent token selection can produce inconsistent continuations that shorten the accepted prefix. Existing methods mostly leave conditional decoding to a lightweight module after the backbone, which limits the flow of predecessor information to successors. Our analysis of DFlash shows that early positions already form recoverable predictions in shallow layers, and that accurate adjacent predecessors help successors more when they enter earlier. We therefore propose DSpine, a drafter with causal conditioning injection throughout the backbone: at every layer, gated adjacent injection writes each predecessor's predicted feature into its successor, so the causal conditioning chain unfolds over network depth while all positions update in parallel. A unified transfer space built from the target model's output embeddings unifies layer-wise injection with predecessor-conditioned decoding, and layer-wise output-embedding supervision promotes the formation of predicted features in shallow layers. Fused kernels and a transition cache execute both efficiently in parallel within SGLang. Across seven math, code, and chat benchmarks, DSpine achieves the longest acceptance length at both temperatures on Qwen3-4B and Qwen3-8B. At temperature zero on Qwen3-8B, it raises the seven-benchmark mean from DFlash's 3.77 to 4.82 (+27.8%); in SGLang serving tests, it delivers 23.3% higher throughput than DFlash on average.

Group Relative Policy Optimization (GRPO) has become a promising approach for training large language model agents. However, its uniform assignment of trajectory-level advantages to all policy tokens fails to distinguish consequential decisions from less relevant ones, obscuring which intermediate decisions contributed to success. We introduce ProVer, a framework that targets potentially pivotal decisions for fine-grained credit assignment in agentic reinforcement learning. Given a rollout group, an agentic judge contrasts successful and failed trajectories to propose a segment potentially responsible for their divergent outcomes. Rather than directly trusting the judge's assessment, ProVer verifies the proposed segment by estimating its advantage from the difference in terminal success rates between current-policy continuations sampled before and after the segment. Positive estimates are then incorporated into the GRPO advantages of policy tokens within the proposed segment. By using model judgment only to select where to verify, ProVer grounds local credit in observed outcomes without exhaustively evaluating every intermediate state. Across ALFWorld, WebShop, and SearchQA, ProVer achieves the strongest average performance at both model scales, with relative improvements over GRPO of 9.91% and 7.12% for Qwen3.5-2B and Qwen3.5-4B, respectively. Further analyses demonstrate that informed segment selection improves policy training with modest additional generation overhead, even without a frontier-scale judge model, highlighting the effectiveness and efficiency of selectively targeting pivotal decisions for fine-grained credit assignment in agentic reinforcement learning.

Computer-use agents (CUAs), while capable of completing computer tasks in everyday and professional workflows, can cause unintended harm even under benign instructions and environments. However, detecting such harm remains challenging. First, it requires careful, task-specific reasoning: verifiers guided only by general safety criteria often overlook many important but subtle harmful behaviors. Second, it requires active investigation: past trajectory screenshots show what the agent did but not always what actually changed in the environment, so LLM-as-a-judge verifiers that rely on screenshots alone may be unable to determine the actual consequences of actions. To address these challenges, we introduce SCOUT, a two-stage agentic safety verifier that synergizes reasoning-intensive rubric generation with tool-intensive evidence gathering. First, our SCOUT rubric generator extensively reasons over the task and the agent's trajectory to determine what successful and safe execution should entail, generating task-specific completion and safety rubrics. Then, our SCOUT probing agent follows these rubrics to interact with the post-execution environment and collect grounded evidence for final safety and completion judgments. We evaluate our framework on two computer-use safety benchmarks. On AutoElicit-Bench, SCOUT achieves 75.4 unsafe F1 and 74.5 completion F1, outperforming LLM-as-a-judge verifiers and naive tool-use verifiers. SCOUT leads on OS-Blind with 76.4% unsafe detection accuracy. Test-time reflection reduces final unsafe execution rates from 30.2% to 17.2% on AutoElicit-Bench. Ablations and analysis show that tool-free rubric generation in SCOUT elicits substantially more reasoning and is crucial for safety detection across verifier backbones, especially non-frontier ones. A preliminary extension to coding tasks shows that SCOUT can support safety verification beyond computer-use.

Input encodings can restrict which measured contrasts a predictor can jointly reproduce, even when no single contrast is forced to vanish. We compute the attainable contrast space from an encoder's equivalence classes and a fixed contrast design, without labels, loss, or a fitted model; projecting the recorded contrasts onto that space gives an empirical error floor for any unrestricted decoder on those classes. On a 140-rectangle siRNA interaction panel, a graph neural network's training-only feature mask merges 165 endpoint states into 90 classes and cuts the rank of the 140 interaction contrasts to 72. The resulting floor is 0.009980, which is 14.6% of the fitted model's interaction squared error; the fitted model reaches 0.068335, slightly worse than a control predicting no interaction at all. A minimum of three restored chemistry columns recovers full rank. Refitting without the mask removes the floor entirely, yet interaction MSE improves by only 0.000017 under the reported protocol, and the restored columns remain absent from every training input. On a released RNA-splicing predictor, whose encoding is injective on the measured states, the same computation returns the full design rank of 1,986 and a floor of exactly zero. These results separate what an encoding permits from what a fitted model achieves; they do not identify what limits the remaining error. The rank check needs no fits and bounds what any amount of training under a fixed encoding can recover. The project repository is available at https://github.com/shadi97kh/REPRESENTABLE-BUT-UNLEARNED.

Large language models are increasingly evaluated in specialized domains such as law, medicine, software engineering, and cybersecurity, yet film remains comparatively underexplored despite requiring long-form narrative integration, multilingual interpretation, and culturally situated audience judgments. We introduce CineSubBench, a benchmark for evaluating long-context film understanding from multilingual movie subtitles. A subtitle track represents a film as thousands of short, temporally ordered utterances from which models must reconstruct characters, relationships, events, causal progression, and themes without explicit scene or event structure. CineSubBench contains 1,012 films with complete subtitle coverage in six languages, yielding 6,072 tracks and 8.13M timestamped subtitle entries. It provides a matched multi-task, multilingual, and multicultural (MultiX) evaluation setting: seven tasks span narrative reconstruction and abstraction, genre prediction, age suitability, country-specific motion-picture ratings across ten national classification systems, and subtitle-grounded language safety. Across nine LLMs, plot premises are recovered more reliably than event-complete synopses; cross-lingual consistency varies substantially across models and languages; national rating systems expose distinct calibration patterns; and strong profanity is far easier to ground than mild obscenity. CineSubBench establishes film as a long-context LLM evaluation domain and provides a unified benchmark for measuring narrative, multilingual, cultural, and evidence-grounding capabilities.

Large language models (LLMs) show considerable potential in simulating human attitudes and preferences. Prior work finds that LLM-generated responses can compress the range of attitudes found within populations and misrepresent particular subgroups in ways that vary across models and topics. We introduce Population Fidelity, an evaluation framework that distinguishes key conditions required for a set of LLM-generated responses to represent a population. It incorporates three dimensions: group-level accuracy, the amount of between-group variation, and the structure of that variation. We demonstrate the framework's utility in two ways. First, we reproduce a prior study of "machine bias" in LLM survey responses and apply the framework to its models and more recent ones, showing that poor representation reflects not only insufficient between-group variation but also variation assigned to the wrong groups. Second, we evaluate one proposed approach to improving models' population representativeness: cultural fine-tuning. We find that cultural fine-tuning can improve alignment with the survey center without improving the representation of within-population differences, a distinction that measures of aggregate agreement do not capture. We argue that representing a population requires models to reproduce several features of human attitudinal variation simultaneously. Our framework organizes these features and provides reusable code, data, and trained models for evaluating population fidelity across substantive domains and assessing proposed alignment methods.

Existing multimodal scaling laws fit multimodality terms empirically after testing and never vary how much data is multimodally paired at fixed data budgets. We investigate how, under the same total data per modality, changing the number of paired data affects loss curves in multimodal classification tasks. We train models in three different environments and run experiment sweeps varying data sizes and pairing budget. Pairing ratios have a dramatic impact on loss and this impact is directly tied to how much information synergy the task contains. Only paired data is able to reduce synergistic loss, while unpaired data can reduce redundant or unimodal information up until unimodal floors. Unlike traditional scaling laws where loss drops immediately in power law decay, synergy acquisition is gated, requiring a critical threshold of paired data before synergistic loss falls at all. We introduce a new family of multimodal scaling laws where total data-attributable loss is the sum of four individual power laws corresponding to the four different information channels of redundancy, a unique channel per modality, and synergy, and show how this law is both more theoretically sound and empirically valid across our experiments. This law predicts multimodal loss in our experiments more accurately than existing laws, with 3.2% error on fit tests versus 10.4% error for the best pairing extension of published laws.

Large language models can learn abstract rules from just a few in-context examples, but how their internal mechanisms activate as examples accumulate is not well understood. We trace a three-stage symbolic reasoning circuit (abstraction, induction, retrieval) across shot counts in three model families and find it is detectable and functional well before the model achieves high accuracy. Per-head causal contribution grows up to 8x from 1- to 10-shot, and cross-shot activation patching raises accuracy from 1% to 56% at 0-shot and 17% to 88% at 1-shot. Function vectors scaled and injected at 0-shot rescue accuracy up to 86%, largely substituting for the induction stage but depending critically on an intact downstream retrieval stage. The infrastructure for abstract rule-following is present in the weights before any demonstrations; in-context examples, function vectors, and related interventions appear to supply input to the same latent circuit.

The shallow embedding of higher-order modal logic in classical higher-order logic, used in Benzm\"uller and Scott's Notes on G\"odel's and Scott's variants of the ontological argument (2025), reaches beyond the modal object language of the arguments: its property quantifiers range over terms that may also express nominals and satisfaction operators of hybrid logic, and a proof using one proves a theorem of the embedding that need not be one of the modal logic. That the framework affords this is not new, and whether a result is one of the modal logic can be settled in two ways: by replaying it in an explicit proof calculus, done by hand for chosen theorems, or by analysing the proofs the embedding itself produces, which this article does mechanically, for every result at once. Every statement the Notes prove has a proof inside the object language: 294 written out by hand and machine-checked, none using a nominal. The proofs the Notes themselves give instantiate no nominal either; what the detector flags there are terms a prover substituted. The three questions the Notes leave open are settled too, and without nominals, but the conjunction axiom has to be emended: generalised in the Notes to G\"odel's "any number of summands", it covers the conjunction of no properties, and of one; the empty one alone settles all three, and the two together yield what a separate axiom of G\"odel's is for. This article restricts the conjunction axiom to at least two different conjuncts, the reading G\"odel's footnote suggests, and the questions are settled again, by proofs that turn on the argument rather than a degenerate instance. The restriction holds of the object language only: with a nominal the axioms make the accessibility relation the identity and the readings coincide. Every theorem is verified in Isabelle/HOL and independently in Lean 4; the countermodels are Nitpick's, certified by the build.

Prompts are the primary mechanism for directing the behavior of large language models (LLMs). Yet the internal structure and causal hierarchy of prompts remain poorly understood: which parts are causally necessary and which are redundant is an open question. This opacity can have severe consequences. Subtle prompt variations can silently shift model outputs in critical software systems, and engineers lack techniques to reason about prompt reliability. We present \framework, a blackbox prompt-minimization framework that reduces few-shot prompts to their necessary minimal subset. We use a case study to apply \framework to a few-shot learning system and demonstrate the insights that this framework can provide. Our experiments show that few-shot exemplars can be reduced by a mean of 65.3\%~$\pm$~15.8\% in character count while fully preserving propositional output fidelity. The models preferentially retain logical identifiers and constraint declarations while discarding natural language prose and cross-prompt relational annotations. Our analysis also shows that some models are universal encoders, able to produce highly legible yet minimized prompts, while others are universal decoders, able to interpret minimized prompts from most other models. By identifying which components are indispensable, \framework provides a principled basis for prompt compression and structural analysis of few-shot exemplars.

Mixture-of-Experts (MoE) layers are central to frontier language models, and recent architectures push toward more and smaller experts. In this regime, the standard linear router becomes a bottleneck: with $M$ experts and hidden dimension $h$, its per-token cost $\Theta(Mh)$ dominates the MoE layer once $M$ is large. We introduce MoRE (Mixture of Rank-reduced-routed Experts), which factorizes the router weight matrix at rank $r$ and reduces the routing cost to $O((h + M)r)$. We prove that rank logarithmic in $M$ suffices for routing expressivity when the number of active experts is fixed, and is necessary up to precision factors. We also prove that logarithmic rank preserves load balance in a Gaussian memorization model, and training on a synthetic phonebook task shows that low rank does not hurt memorization. At matched active FLOPs, the factorization allows a factor of $\Theta(h/r)$ more experts. To realize this gain in wall-clock time, we design a fused Triton kernel at inference that avoids expensive memory operations on HBM. Empirically, MoRE improves memorization on the phonebook task and performance on knowledge-intensive Q\&A benchmarks after pretraining, while matching reasoning ability. Code available at https://github.com/Matheart/MoRE_code.

Evaluation awareness (EA) can cause large language models (LLMs) to behave differently during audits than in deployment, yet measuring and accounting for EA remains challenging. We introduce verbalization training (VT), a method for making LLMs less reticent about verbalizing evaluation awareness while avoiding to supervise the latent belief itself. VT uses a model's spontaneous verbalizations as evidence that awareness is present and truncates each rollout immediately before the verbalization, producing training prefixes at which the model is presumed to be aware. The model is then trained with an RL objective designed to increase verbalization in a calibrated way. Across Qwen3.6-35B-A3B, Kimi K2.6, and Inkling, VT increases verbalized EA by 2.4-2.9 times and transfers to held-out agentic settings, while measured latent EA and behavior remain largely stable. In a causal experiment, we independently implant meta-knowledge about evaluations through synthetic-document fine-tuning and show that VT-induced verbalizations reflect the richer knowledge acquired by the model.

Chunked KV-cache compression reduces the memory and attention costs of long-context inference by compressing windows of consecutive tokens into fewer cache entries at a fixed stride. Such compression also introduces a new positional coordinate: a token's phase, or its position relative to compression-window boundaries. We uncover a systematic asymmetry in models using such compression: the same information can be easy to retrieve at one phase and difficult at another. We call this periodic variation in retrieval performance phase sensitivity. In large open-weight models with such compression, long-context retrieval accuracy can differ by up to 40 percentage points across phases, revealing periodic weak spots that average benchmark scores can conceal. To investigate this behavior, we pretrain a family of transformers from scratch across multiple KV-compression designs, reproducing phase sensitivity across the variants. Mechanistic analysis using causal interventions in these models reveals phase specialization: different attention components contribute asymmetrically to retrieving information at different source phases. We further analyze idealized retrieval models, showing how gradient flow dynamics may favor sharp phase specialization. Evaluating models with chunked KV-cache compression thus requires measuring across compression phases: high average accuracy can coexist with systematic positional failures.

Decentralized Federated Learning (DFL) eliminates the central aggregation server, reducing the single point of observation that traditional defenses against attacks rely on. As a result, peer-to-peer networks become exposed to malicious updates containing backdoors or semantic poisoning, since such updates can remain close to benign ones in the parameter space while behaving very differently. This may evade defenses based on passive parameter inspection. However, existing deception-based defenses have mainly been designed for centralized FL and do not jointly address local observation, poisoning propagation, source attribution, and containment in strictly serverless DFL. To address these limitations, this paper presents DecoyTrace, a proactive cyber deception-based defense for strictly serverless DFL environments. DecoyTrace deploys a mobile DecoyNode that generates decoy challenges using chaotic maps, disseminates a dual model (clean vs. decoy) based on neighbor trust, and evaluates them using three-state semantic metrics. Upon confirmation, a distributed protocol isolates the source and performs a model reset or recovery to preserve training progress. Evaluated across sixty configurations on the NEBULA platform (five datasets, three topologies, and four attack/defense scenarios), DecoyTrace systematically restores lost utility. The F1-score remains within 0.03 of the baseline on MNIST/FashionMNIST (mitigating drops of up to 0.37), matches or exceeds the baseline on EMNIST and CIFAR-100, and remains between 0.05 and 0.10 below the baseline on CIFAR-10, the most visually complex convolutional scenario evaluated. Furthermore, containment reduces CPU and network usage by up to two-thirds. These results demonstrate the feasibility of unifying deception, identification, and containment in DFL, while also identifying its limitations in complex tasks and multi-attractor threat models.

The state-of-the-art Membership Inference (MI) methods calibrate their signal separately for each example using reference models, auxiliary models trained to exclude the target. This paradigm scales poorly to modern large models, however, whose training is too expensive to replicate. This has motivated one-round settings, where only a single trained model is available; but without reference models the per-example calibration that drives the strongest attacks can no longer be estimated, leaving the membership signal weak. We ask whether neighbors of the target point can recover this calibration without training any additional model. Our key observation is that reference models serve only to reveal how an example behaves under models not trained on it, and that querying the target model on nearby samples yields the same information. We propose two complementary ways to obtain such neighbors, and show that querying them against an early training checkpoint further sharpens the signal. We evaluate across three image classification datasets and three training setups, showing that neighbors yield strong membership signals and competitive attack performance at no additional training cost.

Latent world models rely on representation geometry for planning, yet regularizing the latent marginal alone does not determine the state-to-state relationships used for action selection. We show that this can cause planning-relevant novelty structure to be weakened as representations are transformed into the final latent used by the planner. We introduce Aligned Transport of Latent Structure (ATLAS), a training objective that explicitly preserves relational geometry while calibrating the global latent distribution. ATLAS transfers normalized pairwise structure from an informative encoder representation to the planning latent and uses Wasserstein embedding matching (WEMReg) to calibrate its marginal through one-dimensional Wasserstein-2 transport. Our analysis shows that relational preservation and marginal calibration impose non-redundant constraints, and connects finite-candidate planning stability to relational distortion, latent-scale mismatch, and prediction error. Instantiated in LeWM, ATLAS improves mean goal-reaching success across PushT, TwoRoom, and OGBench-Cube on both lower- and higher-novelty evaluation subsets, with the largest gain on higher-novelty TwoRoom episodes. Representation and rollout diagnostics further show stronger novelty-related structure in the planning latent, improved marginal calibration, and lower multi-step prediction error. Together, these results highlight preservation of planning-relevant latent geometry as an important ingredient for reliable world-model planning. Code is available at https://anonymous.4open.science/r/atlas-world-model-72C4/.

Deep learning models have achieved strong performance in artificial intelligence for science, yet their black-box nature limits our understanding of how they learn scientific tasks. Existing methods for interpretability provide limited insight into how models organize evidence and evolve during learning. We introduce explainability from training (EFT), a model-agnostic paradigm that traces model interpretation during training to explain why models rely on specific features and how they organize these features as predictive evidence. We apply EFT to four state-of-the-art T cell receptor (TCR)-epitope prediction models, TCR-SRIM, TULIP, MixTCRpred, and NetTCR-2.2, spanning post-hoc and interpret-by-design approaches as well as transformers and CNNs. To investigate how structural information affects model explanations, we introduce a benchmark, TCR-XAI2, containing 388 unique experimentally resolved TCR-epitope structures, complemented by structures predicted using AlphaFold3, Boltz-2, TCRModel2, tFold-TCR, and OpenFold3. Using EFT with TCR-XAI2, we demonstrate that (1) CNN and transformer models exhibit distinct learning trajectories; (2) TCR $\alpha$ and $\beta$ evidence can conflict during learning, limiting the benefits of jointly modeling both chains, while MHC information mitigates this; and (3) real versus predicted structural data for TCR-epitope prediction exhibits distinct TCR and peptide feature preferences as well as differing trajectories of model certainty.

Understanding how the brain parses actions and events from time-varying natural inputs is a central challenge in neuroscience. Recent work has used deep neural network (DNN) models to build stimulus-computable fMRI encoding models that predict single-voxel responses to complex natural videos. However, the majority of video-computable encoding models predict responses using simple linear mappings from model tokens, overlooking the spatiotemporal structure shared by video representations and neural responses. Recent cross-attention encoding models address this limitation for static images, enabling flexible stimulus-dependent weighting of image content across space. Here, we extend this framework to naturalistic video, using per-parcel cross-attention to dynamically route features from a self-supervised video model (V-JEPA-2) across both space and time, fitting this model to fMRI responses to short video clips. We compare joint spatiotemporal attention with factorized and selectively constrained alternatives, and find that joint routing improves predictions of brain responses to held-out videos across higher visual regions, most consistently in lateral and dorsal visual areas associated with dynamic motion perception. Moreover, our method provides interpretable, stimulus-specific attention maps that dynamically follow moving objects, revealing which locations and temporal moments contribute to each neural response. We further show that attention maps from parcels in different category-selective networks (face-, body-, scene-selective) differentially weight content in accordance with expected semantic selectivity. Together, this work provides a new computational framework for understanding how visual information is adaptively weighted by cortical populations during dynamic visual perception.

Test-time scaling improves software engineering agents by generating multiple candidate trajectories and selecting the best one. Verifying and selecting among these long interactions can consume as many tokens as generation itself. Existing hybrid workflows first apply an LLM-based execution-free (EF) verifier to filter candidates before running tests, which adds another model pass over every trajectory. We introduce LatentSift, a token-free and execution-free filter that replaces this first stage with hidden states the policy already produces while generating the candidates. It represents each candidate through its reasoning, observation, and function-call states, compares them with positive and negative banks of such states collected from successful and unsuccessful trajectories during policy training, and fuses the resulting distance scores with a learned linear score to retain promising candidates for the execution-based stages. On SWE-bench Verified, across three agents and two policy sizes, LatentSift cuts EF-verifier tokens by 66.6--81.0% and total verification tokens, which include test generation, by 49.1--62.1% at K=16, while hybrid Best@16 matches or improves on each agent's reference workflow, rising from 59.26% to 60.06% on DeepSWE-Preview.

A personal language agent that acts for its owner across private and shared conversations can learn a fact from one audience and later place it in the context it assembles for another. We study authorization before context across the whole memory lifecycle. Each memory item carries the audience present when it was recorded; derived items are partitioned by audience, receive the intersection of their sources' audiences, or are suppressed; an audience widens only by an explicit, object-specific grant; and an item enters a model attempt only when every current viewer belongs to one of its authorized audiences, with unresolved viewers failing closed to public-only. Under explicit identity, provenance and complete-mediation assumptions, this admission is sound and policy-complete on the exact assembled context, enforced by exclusion rather than by model behavior. We realize it in two independently persisted reference architectures, a flat store and a relationship graph, and, descriptively, in a native agent-memory runtime. In a prospectively frozen confirmation over 10,000 multi-party histories, no forbidden item entered any architecture's context, whereas unscoped retrieval exposed forbidden items in 82% of its contexts. Entitled recall matched policy-equivalent baselines exactly and exceeded unscoped retrieval by 0.30 Recall@5, with a Holm-confirmed advantage that grows with distractors. No architecture produced a wrong-principal substitution, but unscoped substitutions were too rare to establish the prespecified joint decision.

Dense retrieval, the key component of Retrieval Augmented Generation (RAG), retrieves the most relevant documents by comparing dense vector representations of queries and passages from a large corpus. In privacy-sensitive applications, the server observes the query and controls which evidence is returned, creating both confidentiality and integrity risks. We formulate private dense retrieval as providing query privacy and retrieval integrity against a malicious server, and develop a two-round cryptographic protocol that provides both guarantees. Our protocol reduces private and verifiable retrieval to multiplication of a committed matrix by an encrypted vector and uses low-bit quantization to make this computation practical. We evaluate the resulting trade-off between cryptographic cost, retrieval quality, and downstream RAG accuracy across six embedding models, four language models, and corpora of up to 2.68 million passages. Our results show that, with a clipped quantizer, three-bit quantization largely preserves retrieval quality and downstream accuracy, while a private query over a corpus the size of a clinical reference requires one to three minutes of server time. These results suggest that private dense retrieval is already practical for moderately sized, privacy-sensitive corpora when minute-scale latency is acceptable.

Traditional reinforcement learning (RL) techniques focus on maximizing expected cumulative reward, where each action assumes to take a constant unit of time. However, this assumption does not hold for agentic RL tasks such as machine learning engineering (MLE) agents, where actions involve data loading, feature engineering, and model training that take variable durations. Efficiency matters in modern agentic RL where actions are costly. To address this limitation, we adapt from continuous-time RL and Semi-Markov Decision Process (SMDP) formulation and propose Reward-rate Policy Gradient (RPG), where we focus on optimizing the reward rate -- the long-term reward per unit of time. RPG estimates the reward rate from off-policy samples, then charges each action for the time it consumes at that rate. We first conduct theoretical analysis in the bandit setting to establish that RPG approximates the optimal reward rate and empirically demonstrate it outperforms baselines while avoiding enumeration over the policy space, a known issue for an existing method. We then further apply RPG on a small language model (Qwen3.5-4B) with self-improvement loops and empirically show it obtains higher rewards within a fixed time budget than vanilla RL on MLE-Bench and NanoGPT, with a 19.2% and 85.7% margin, respectively. Our method provides a practical solution for optimizing performance under wait time considerations in modern agentic RL tasks, where actions interact with external environments and cost time.

Generative models make molecular crystal structure prediction fast, but their samples still exhibit geometric and packing violations. Physics can be introduced during training, post-training, or inference, yet these choices are rarely compared with the generator and physical signal held fixed. We introduce CrystAF, an all-atom crystal flow-map generation model, and use it with the UMA interatomic potential to systematically study where physics should enter. Post-training learns physical preferences directly into CrystAF, improving molecular validity and crystal packing while leaving sampling unchanged: physics is paid for once during training rather than repeatedly at deployment. In contrast, UMA relaxation is effective at repairing local clashes but makes generation 6--26$\times$ slower, while learning from relaxed targets provides little benefit. These routes are complementary rather than competing. Physics-informed post-training first shifts the generated distribution toward more physically reasonable structures, after which inexpensive inference-time corrections further remove clashes and restore stereochemistry that the generator cannot represent. Importantly, the same post-training strategy also improves the multi-step all-atom Clari-M and rigid-body MolCrystalFlow generators, demonstrating transfer across architectures and representations. Together, our results suggest a simple principle: learn reusable physical alignment into the generator, and reserve inference-time physics for residual constraints that are better corrected than learned.

Decision-only language models return a probability for every answer option instead of generating text, which makes them attractive as survey respondents and as judges. We audit the cultural values of one such model, TypeSafe's JEV, with the Values Survey Module 2013. We asked it the 24 items as 12 matched Saudi and 12 matched American personas and without a persona, in English and Arabic, under eight ways of formulating the request (288,000 answers). JEV's answers were highly repeatable (ICC 0.997), and without a persona they resembled those of its own American personas. When the persona was Saudi rather than American, the answers moved in the direction of the human Saudi-US difference, reproducing 87% of its size in English but 62% in Arabic, with long-term orientation reversed. A language cross shows that the smaller difference in Arabic comes from the language of the items, not from the language of the persona description. Age shifted the profiles about as much as nationality, gender shifted them more for Saudi than for American personas, and JEV was less confident in Arabic and for Saudi personas. These patterns held in every request design, although the model never generates text.

Longitudinal data are valuable because people change. Yet the objectives used to learn from these data can inadvertently erase that change. In person-level contrastive learning, observations from the same person are treated as positives; as records grow, those positives can span increasingly distant---and increasingly different---behavioral states. More history can therefore produce not only more data, but broader invariance. We show that this distinction is fundamental. We separate \emph{record span}, how much history the learner sees, from \emph{supervision span}, how far across that history positive-pair supervision reaches. Across in-home sensing records spanning up to 2.7 years, broader supervision systematically suppresses recoverable changing-state information, even when the available history is held fixed. At the broadest span, less than 10\% of the information recoverable from an untrained encoder remains. Yet keeping positives local is not sufficient: as records grow, even distant states that are never paired become increasingly similar. Explicitly contrasting other observations from the same person reverses this loss without shortening the record, revealing a second route by which longitudinal scale can broaden invariance. Finally, we prospectively reproduce the supervision-span effect in 199 GLOBEM participants. Longitudinal scale therefore presents a choice: more history need not mean more invariance. By controlling what is held invariant as records grow, we can preserve the change that made the longitudinal data valuable in the first place.

Masked diffusion language models (MDLMs) can generate text efficiently by predicting multiple masked tokens in parallel, but predictions from the same forward pass are not necessarily reliable when committed together. We study when parallel commitment is reliable. Our diagnostics show that confidence alone does not determine a reliable commitment order: confident predictions near the end of the sequence can fix an answer before its supporting computations are established, and downstream predictions become less reliable as the uncertainty of their upstream context grows. At the same time, a single forward pass can already resolve several masked tokens, and predictions that remain stable across the final layers are more likely to be correct. Based on these findings, we propose Reliable Parallel Decoding (RPD), a training-free method that selects candidates by layerwise prediction stability and final confidence, and commits them under a cumulative entropy budget over their preceding masked positions. RPD defers predictions with uncertain upstream context while committing the remaining candidates in parallel, without relying on a fixed block schedule. Across mathematical reasoning and code generation benchmarks on LLaDA and Dream, RPD achieves the highest decoding throughput among the evaluated methods while maintaining or improving accuracy.

Multivariate time series forecasting (MTSF) is critical across many real-world domains. Existing deep learning approaches fall into two paradigms with distinct limitations: channel-independent (CI) methods unconditionally ignore cross-variable dependencies and model only temporal dynamics, while channel-dependent (CD) methods consider both but typically rely on architectural compromises to mitigate overfitting and computational overhead. We therefore propose Chameleon, a specialized CD state space model (SSM) that enables data-dependent, fine-grained interactions across variables while scaling linearly with their number. By connecting selective SSMs with the Kalman filter, we leverage the missing measurement update in the former for cross-variable modeling while preserving the SSM backbone for robust temporal modeling. We further identify favorable inductive biases of GatedDeltaNet for time series, adapt it as our backbone, and improve generalization through additional techniques, including a previously unexplored stochastic perturbation of reversible instance normalization. On strongly dependent ODE and PEMS datasets, Chameleon achieves the best MSE and MAE across all settings, while its CI ablation and prior CD methods incur 61-178% higher MSE on average. Across 28 standard benchmark settings, Chameleon also achieves better MSE and MAE than each baseline in at least 27 and 22 cases, respectively. Training-time and peak-memory analyses on Traffic and ETT further demonstrate competitive efficiency and favorable memory scalability across different variable counts.

Several tonal pitch spaces and computational models have been proposed to analyze tonal structure in Western tonal music, many of them grounded in principles from music theory and used to support tonal analysis with important implications for tonal tension. In parallel, data-driven methods such as skip-gram have been used to learn chord embeddings from symbolic corpora, but their ability to recover tonal structure and its relation to tonal tension remains underexplored. In this work, we investigate how skip-gram chord embeddings reflect tonal structure and whether they provide a useful basis for analyzing structural aspects of tonal tension. Using chord sequences with and without transposition-based augmentation, we evaluate the learned spaces from geometric, functional, and tension-related perspectives. We show that augmented embeddings exhibit strong transposition equivariance, recover a clear circle-of-fifths structure, and support interpretable shifts between key-related regions of the learned space. We then derive embedding-based measures from chord-to-key distance and contextual chord-distance relations, and show that they capture meaningful aspects of tonal tension structure through correspondence with matched tonal measures and moderate alignment with human tension profiles. Across analyses, transposition-based augmentation generally improves the stability, tonal coherence, and interpretability of the learned space.

Guard models are used as safety classifiers, with confidence scores driving downstream moderation decisions. We evaluate five guard models for prompt classification and find that although several are nearly calibrated on clean inputs, adversarial attacks degrade their calibration by an order of magnitude, turning false negatives into high-confidence errors indistinguishable from correct detections. Comparing each guard with its corresponding base LM, we find that uncertainty signals often remain available, with the base model typically expressing uncertainty on the same inputs where the guard fails. Layer-wise analyses localize this guard-base divergence to later layers, where guard models exhibit sharper safe/unsafe separation and lower-rank representations, while adversarial harmful inputs lie closer to the clean-safe region. These findings highlight a mismatch between guard confidence and base model uncertainty under attack.

Quantum computing has already been explored in several network-security applications. However, how quantum computing may contribute to network-security classification in both the near term and the longer term has not been systematically discussed. This paper studies this question through two complementary experiments. First, we evaluate near-term quantum-kernel support vector machines (SVMs) on practical network-security classification tasks and compare them with classical SVM baselines on KDD Cup 1999, CICIDS2017, and BoT-IoT. Across these runs, quantum kernels are competitive. They can match or improve classical baselines in some settings, while classical RBF kernels remain stronger in others. This suggests that near-term quantum-kernel methods should be evaluated as practical, dataset-dependent alternatives to classical kernels rather than as uniformly superior replacements. Second, we use quantum oracle sketching (QOS) to study a longer-term memory advantage for classification with streaming classical samples. In QOS, samples are processed online and used to incrementally construct an approximate quantum oracle, which provides coherent query access for downstream quantum algorithms without retaining the entire dataset. Under the QOS-inspired machine-size estimate, comparable accuracy corresponds to a substantially smaller effective memory-size proxy than explicit sparse/QRAM-style storage. Compared with a simple streaming proxy, the result is more nuanced because aggressive feature filtering can make the streaming dimension small. This suggests that the long-term value of quantum computing for network-security classification may lie in memory-efficient data access rather than immediate runtime speedup. Together, these experiments show how quantum computing may contribute to network-security classification from near-term classification performance and longer-term memory efficiency.

Latent space monitors aim to detect undesired behaviors in LLM agents by inspecting an agent's internal activations rather than its outputs. However, interactive monitoring creates a feedback channel where each verdict the monitor delivers leaks information to the model about how its internal states are being evaluated. We ask whether an agent can infer the monitor's decision rule from this feedback and then selectively edit its activations to evade detection. Unlike prior evasion attacks, the model is never explicitly told what the monitor detects. Surprisingly, off-the-shelf models already produce activation edits aligned with the monitored direction, but at insufficient magnitude for evasion. Simply scaling up these edits by a factor of 8 reduces the monitor's TPR from 100% to 27%. A rank-1 LoRA amplifies this behavior into effective evasion within the forward pass, reducing TPR further to 4% on held-out concept monitors while leaving other concepts at their normal detection rates. Capabilities on standard benchmarks are retained under this finetuning, and the evasion skill survives retraining the monitors on the new activations. Mechanistically, we find evidence that the model computes its activation edit from the prior in-context turns, and show that the edit becomes more aligned with the monitored direction as more examples are provided. These results demonstrate feedback-conditioned control over activations and suggest that latent monitoring should be treated as an interactive process in which agents can observe and respond to oversight measures.

Speaker verification systems encounter combinations of noise, channel distortion, and changes in speech. Evaluating each condition separately does not establish whether their effects add. InterBias-SV organises this question around a four-term comparison: joint error, two marginal errors, and a common reference. Its results artefact contains 4,068 scored records across 17 experiments, 12 encoder labels, and six speech corpora, totalling 12 million trial evaluations. Three experiment families contain the same-corpus terms needed to compute additive contrasts. For labels assigned to speaker-trained encoders, their mean contrasts are +0.0026, +0.0088, and +0.0024 in equal error rate (EER), with larger variation across settings. These descriptive averages do not establish equivalence to additivity: trial matching, checkpoint identity, and parts of the condition metadata remain unverified. We also examine two interpretation problems. Near-chance EER can make additive predictions difficult to interpret, but chance performance is not a hard EER ceiling, and correlation with the prediction does not identify a saturation mechanism. Ratios of demographic gaps are unstable when their clean reference is near zero; absolute gaps provide a more direct summary. The benchmark provides condition definitions, analysis scripts, and explicit requirements for interpretable compound-condition comparisons, while separating recomputable summaries from claims that require further experimental validation.

Past research using log data has faced the "learning system wall," whereby few methods exist for generalizing models of student learning across platforms. Increasingly, online learning is captured by richer forms of data, including dialog and video, with new affordances. An example of this is remote tutoring programs, where human tutors support students who use learning systems while video conferencing. Toward better platform-general modeling of learning, we introduce an AI-driven multimodal transcription system that processes screen-recording videos into unified screenplay-style transcripts containing audio dialogue and annotated learning log actions. We describe a planned method for temporally aligning AI-generated multimodal transcripts with MATHia learning logs and for identifying and classifying student learning processes to align with MATHia logs. Lastly, we highlight challenges and potential solutions in capturing learning processes in one system, offering initial steps towards generalizing log data across diverse systems.

A common assumption in language model development is that cognitive abilities are organized around a general, domain-free intelligence factor, like fluid intelligence in humans. This assumption is rarely tested directly, and prior attempts have done so only at a much smaller scale. We take a latent variable approach to intelligence in language models, similar to how psychometricians study psychological constructs. Performance in every specific problem set is influenced by a domain-specific and a domain-agnostic latent factor. Using factor analysis as a dimension-reduction technique, we analyzed 13,251 published evaluation scores covering 1,618 language models across 456 different text-only benchmarks. Due to the super-sparse nature of the dataset, we triangulate our analysis across different data densifiers and imputation methods. A robust pattern across different modes of bias is that 1. A general intelligence factor accounts for 70.8% of variance in model performance at our most generous estimate, and far less than that in most of our solutions, 2. Content-similar benchmarks do not necessarily cluster together, and 3. The $g$ factor is not dominated by any common theme, and there is a lack of evidence that it is well-proxied by standard "intelligence" benchmarks. Our findings go against current endeavors of defining, identifying, and targeting general intelligence as a tangible construct in language model development. This leaves the strategy of targeting a single conceptual ability without support, since the first-order abilities it would have to reach are often partially idiosyncratic and not identifiable in practice.

Robots must often continue a task after a target moves, the viewpoint shifts, or an obstacle appears, even though their earlier observations and committed actions reflect the previous scene. Many simulation robustness benchmarks fix external conditions at reset, leaving this temporal challenge underexamined. We introduce LIBERO-MAX, a benchmark of 8,000 paired cases spanning eight types of changes to geometry, observations, appearance, clutter, and paths. Each pair compares task execution with and without a mid-task event, holding the task, initial state, policy seed, and pre-event action sequence fixed. This controlled comparison distinguishes event-associated regressions from failures already present without the change. Across fourteen current VLA, hybrid, and world-action policies, events reduce success by 11.0-25.7 percentage points. Event profiles reveal shared vulnerabilities to geometry and observation changes, while policy-family rankings interleave. Camera controls show that robustness reflects both competence under the changed conditions and the trajectory from which they are encountered; varying query cadence does not eliminate the gap. Together, the paired protocol and temporal diagnostics establish LIBERO-MAX as a reproducible testbed for diagnosing failures under mid-execution changes and measuring progress toward robot policies that remain effective as the world changes.

Distributed inference has become an indispensable part of deploying medical models under practical latency, memory, and throughput constraints. Although modern frameworks improve serving efficiency through tensor parallelism, mixed precision, kernel fusion, and multi-device communication, they are generally assumed to preserve the behavior observed during centralized HuggingFace evaluation. This assumption creates an evaluation-deployment mismatch: a model may pass offline evaluation but produce a different output after the execution stack changes. To address this mismatch, we propose a testing framework and an improved, distributed-execution-sensitive medical-model benchmark that evaluates the same checkpoint and input under a centralized HuggingFace reference and matched distributed deployments. Extensive experiments across language, vision, and multimodal medical models show that execution changes can produce measurable output disagreements. Across supported visual settings, the test success rate ranges from 0.21 to 0.43 for single-modality models and from 0.32 to 0.98 for multimodal models. The benchmark is aimed at extending medical-model evaluation from capability and security to evaluation-deployment consistency.

Long-horizon agents require context compression to manage growing interaction histories. Compression quality, however, is ultimately determined by downstream execution. Existing prompt-adaptation methods infer compression errors by comparing full-context and compressed trajectories. Such comparisons cannot isolate individual compressions and are confounded by agent stochasticity. We first find that compression degrades reliability before solvability. Using matched counterfactual continuations that compare execution from the same agent state with versus without compression, we further show that severe degradation concentrates at isolated compression events. Motivated by this finding, we propose PAIR (Prompt Adaptation using Interventional Rollouts) for adapting structured compression prompts. PAIR identifies individual compressions that degrade subsequent execution, diagnoses their effects, and revises the relevant sections of a fixed compression template. PAIR achieves the strongest cross-run reliability among compressed methods in every main benchmark-scope combination, consistently exceeding the competing prompt-adaptation baseline. Without modifying the downstream agent, PAIR brings compressed execution close to the no-compression baseline and sometimes numerically exceeds it.

Generative AI has changed how students produce writing assignments. The final artifact is no longer sufficient to understand the process through which it was produced. We introduce DraftTrace, a writing environment that jointly captures three complementary views of writing: the final product, the writing process and interactions with an integrated AI-assistant. DraftTrace reconstructs how a document develops over time and organizes these signals into submission, longitudinal, and class-level analytics for instructors. We deployed DraftTrace in a graduate NLP course with 81 students and compared their sessions with LLM-generated responses entered by automated tools and with copy-typed responses. While product measures distinguish differences in text formulation, process measures distinguish differences in how text is entered. Considering both views together helps characterize cases such as copy-typing. Interaction traces show that students use the assistant differently across stages of writing: to clarify the question at an early stage and to verify answers at a later stage. A preliminary instructor survey highlights the importance of multi-view writing analytics and their interpretability.

On-policy distillation (OPD) trains a student model on its self-generated trajectories with dense token-level teacher feedback. However, naive OPD may suffer from teacher unanchoring, where the student's reasoning trajectory drifts far from the teacher, causing the teacher to be queried on states it would hardly visit and thus provide unreliable supervision. We propose Interactive-Policy Distillation (IPD), which applies adaptive teacher intervention to the student rollout. Under a bidirectional propose-and-verify state machine, the student and teacher alternately exchange their roles as proposer and verifier, and collaboratively generate mixed-source trajectories. Then different supervisions are applied according to the source of each token. This bidirectional propose-and-verify mechanism and the source-split loss make IPD not only a more performant distillation method, but also a unified bridge between on-policy and off-policy paradigms. To make the interleaved dual-model rollouts more efficient, we also design a dedicated fused inference engine that co-hosts both models in one serving instance with separate KV caches and instantiates the state machine model to distribute, collect, and process requests. On math reasoning tasks and across multiple teacher-student model pairs, student models trained with IPD not only outperform those trained with OPD, but also demonstrate higher data efficiency. Specifically, when distilling Qwen3-30B-A3B into Qwen3-1.7B-Base, IPD brings a +3.28 mean@8 and a +3.28 best@8 benchmark-averaged accuracy improvement compared with OPD. Besides, IPD only consumes about 1/4 of the training examples and steps to outperform OPD trained on the whole training dataset for one epoch. We also investigate the impact of different loss variants and takeover / handback configurations, and demonstrate the robustness of IPD on different training data.

Maximum Likelihood Reinforcement Learning (MaxRL) targets prompt-wise log-success and has shown strong performance on reasoning tasks. Under finite rollout budgets, however, the estimator used by MaxRL attenuates each prompt's likelihood gradient by a factor that depends on its success probability and rollout count. Under uniform rollout allocation, the common rollout count fails to compensate for success-dependent attenuation, leaving low-success prompts more strongly attenuated and distorting their relative contributions to the expected aggregate gradient. We introduce SERA (Scale-Equalized Rollout Allocation), which redistributes a fixed rollout budget to approximately equalize these finite-rollout scaling factors. Building on our theoretical analysis of how finite rollouts distort prompt-wise likelihood gradients, we formulate the allocation as a fixed-budget max--min problem, derive a waterline solution to its continuous relaxation, and introduce a multiplicity correction to remove the additional prompt weighting induced by heterogeneous rollout counts. Experiments show stronger alignment with exact likelihood gradients in a controlled ImageNet setting and improved multi-sample solution coverage over MaxRL on maze navigation and mathematical reasoning under matched training rollout budgets.

Extrapolative temporal knowledge graph reasoning (TKGR) predicts future facts from historical snapshots. Most existing methods train once on an early prefix of the timeline and then use a frozen model for all future timestamps. We argue that this fixed-prefix protocol is misaligned with extrapolation. It learns from a static prefix, whereas the target stream is non-stationary: new entities and facts emerge, temporal dependencies shift across regimes, and recurring historical signals must be refreshed online. As a result, models trained only on early snapshots become outdated and degrade over long horizons. We address this mismatch by formulating extrapolative TKGR as continual learning over streaming snapshots. Under this view, effective extrapolation must jointly handle current dynamics, stable knowledge, and recurring historical evidence. Based on these requirements, we propose History-enhanced Two-Step Continual Learning (HiTS-CL), a backbone-agnostic continual learning framework for extrapolative TKGR. HiTS-CL tracks current dynamics via continual fine-tuning, preserves stable knowledge via multi-teacher adaptive distillation, and retains recurring historical evidence via a selective memory of recent and frequent facts. We integrate HiTS-CL into five representative TKGR backbones and evaluate it on four benchmark datasets. HiTS-CL consistently improves extrapolation accuracy, reduces long-horizon degradation, and outperforms strong continual-learning baselines, including a recent method for temporal knowledge graphs. Source code and data are available at https://github.com/liuyansong98/HiTS-CL.

Checkpoint selection is a routine decision in supervised fine-tuning (SFT): training produces multiple checkpoints, but only one is retained. Yet fixed-budget comparisons do not by themselves distinguish three empirical claims: whether more validation data improve checkpoint selection, whether a selection rule outperforms validation-loss selection, and whether it improves over simply retaining the final checkpoint. We therefore treat checkpoint selection as a finite-information decision problem. Holding completed training trajectories, candidate checkpoints, and independent test items fixed, we vary the validation budget and separately measure improvement from additional validation data, gain over negative log-likelihood (NLL) selection, and gain over the final checkpoint. Across 60 mathematical SFT trajectories and 19 configurations, increasing the validation budget from 32 to 305-313 examples raises independent-test accuracy by 0.32 percentage points (pp) for generated-accuracy selection and 0.29 pp for checkpoint agreement, with 95% configuration-bootstrap CIs of [0.10, 0.56] and [0.11, 0.50], respectively. At the full validation budget, the two generation-based rules outperform matched NLL selection by 0.71 and 0.85 pp, respectively, while their gains over the final checkpoint remain unresolved. A cross-domain replication on 12 newly trained Commonsense trajectories shows the same qualitative separation: increasing the validation budget from 32 to 1,024 questions improves generated-accuracy and checkpoint-agreement selection by 0.87 and 0.27 pp, while gains over the final checkpoint again remain unresolved. Together, these results show that benefiting from more validation data, outperforming NLL selection, and outperforming the final checkpoint are distinct empirical claims that require separate evidence.

Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that suppresses this behavior inside the model. Per model, a five-step recipe fits a residual-stream direction from paired episodes differing only in whether an embedded instruction is followed, and retains it only if it passes pre-specified causal and capability gates. At deployment, the direction is subtracted from every tool-result token during prefill. The edit is always on--there is no detection decision to evade--and requires no fine-tuning, auxiliary model, or added tokens, only white-box serving and tool-result span boundaries. Across five open-weights models (8B-106B, five vendor lineages), held-out attack success falls from 0.21-1.00 undefended to 0.00-0.17 defended, and AgentDojo compromise rate from 0.10-0.49 to 0.006-0.079, at 93-100% typography-normalized benign utility, with larger task-dependent costs when reasoning over steered content. A benchmark-level adaptive attacker reaching 0.67-0.73 undefended is held to roughly a quarter of that on the two most deeply evaluated models. Among the defenses we measured on capable models, those achieving lower compromise rates either lost 22-89% of benign utility or fine-tuned the served weights. White-box gradient attacks through the deployed vector compromise at most 2 of 52 episodes, and none of 2,052 replayed human red-team attacks succeeds. CounterSteer largely neutralizes instructional takeover: a black-box framing search cracks 3 of 18 development samples. Parameter manipulation--attacker-chosen arguments in otherwise legitimate calls--is only partially resisted (13 of 18); the decision becomes linearly readable at argument emission but not at the examined pre-generation sites, and is not removed by the tested prefill- or decode-time steering, motivating argument-provenance controls.

Audio large language models (ALLMs) can reason about the content of audio recordings to perform complex tasks. However, these capabilities usually collapse in real-world environments when background noise and competing sources mix the target sound. Inspired by long-term memory in human listening, we propose Long-Term Memory-Guided Audio Enhancement (LTM-AE) to improve selective target perception by refining the audio representations of ALLMs without training. LTM-AE extracts representations in hidden states from separate clean reference recordings as long-term memory for each category, guiding enhancement toward a user-specified listening target. We reconstruct incoming audio tokens in the selected category long-term memory and interpolate the reconstructions with the original tokens before language backbone decoding. This interpolation controls the influence of stored auditory experience while keeping all ALLM parameters fixed. Diagnostic readouts across twenty sound categories and three ALLMs show that LTM-AE strengthens responses to a specified target amid three interfering sources. Averaged over constrained and free-form classification, accuracy gains over raw mixtures range from 29.53 to 46.15 percentage points across multiple open source models. For speech content recovery, LTM-AE with an additional learned token-level gate reduces Qwen2-Audio's word error rate from 23.07% to 14.77%. This work takes an initial step toward using principles of human long-term memory to enhance ALLMs for real-world listening. Our code is available at https://github.com/aynlp/ltm-audio-code

Scheduling problems arise from repeatedly selecting one item from a set of candidates based on their states. These problems often reduce to assigning priority scores and choosing the highest-ranked item. In this work, we propose a factorized scheduling principle (FSP) framework to learn interpretable and transferable scheduling rules. The FSP framework represents system states as condition distributions and decomposes a global scheduling principle into additive univariate and pairwise components with identifiability constraints. The scheduling principle enables the framework to maintain a simple priority-based structure during deployment. This principle is learned by using a policy-based objective combined with a temporal-difference signal defined on the condition distribution. Experiments on synthetic and realistic scheduling tasks demonstrate the FSP framework's strong performance, interpretability, and zero-shot generalization across different system scales.

We study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained coordination skills required for inter-robot collaboration. Supervised fine-tuning (SFT) on multi-robot demonstrations partially bridges this gap, but its performance is bounded by the demonstration data and cannot improve from its own experience. We present a three-stage reinforced fine-tuning (RFT) pipeline for multi-agent VLAs. First, initialization-aware data collection sweeps over initial configurations and invokes human demonstrations only when the pretrained VLA repeatedly fails, yielding robustness to initialization shift with reduced human cost. Second, offline credit-filtered tuning assigns credit to individual agents and fine-tunes on per-agent trajectories with positive advantage rather than on entire joint rollouts. Third, we find existing online RL for VLAs are less effective for hard multi-agent tasks, which we attribute to noisy co-exploration and unstable updates. We instead use online latent-space fine tuning, which freeze the VLA and perform RL in its latent noise space. We evaluate our multi-agent VLA with both $\pi_0$ and $\pi_{0.5}$ backbones across 11 tasks in RoboTwin, RoboFactory and real-world manipulation with two Franka robots. Our multi-agent VLA improves the average success rate by $+23.1\%$, $+16.4\%$, and $+44\%$ on RoboTwin, RoboFactory, and real-world tasks, respectively. Code available at https://anonymous.4open.science/r/mavla_rft-2BC0/.

Creating diverse physical simulations remains labor-intensive because assets, layout, physical parameters, motion, control, and rendering must be designed and debugged jointly. We present Text2Sim, a simulation-specialized agentic pipeline that converts a text-only request into an executable, editable dynamic case. Built on Genesis, Text2Sim uses a hierarchical agentic structure that combines a Planner with specialized Writers, asset-generation tools, and an independent Critic. Compact skills (Debug Cards) distilled from graphics demonstrations provide role-specific physical guidance for execution-based repair. We evaluate physical quality, visual quality, and human preference on 42 held-out prompts spanning rigid, articulated, deformable, and cloth phenomena, with a paper-level split between experience construction and evaluation. We design automatic physical and visual scorers to evaluate the quality of the results, and Text2Sim achieves higher scores than all four state-of-the-art baselines on both metrics. In blinded user studies with these baselines, significantly more participants prefer Text2Sim than prefer the baselines, which is consistent with the results from our automatic scorers. The pipeline also supports a broad range of downstream applications; we select dataset construction and extension to multimodal input as two representative examples. We will release the code, the Debug Card library, and a dataset of generated cases, each pairing the text prompt and rendered video with the executable program, assets, physical parameters, controls, and recorded states.

Visual-memory systems commonly retain or compress past observations. Robot control additionally requires interaction-derived state that no individual frame may explicitly represent, such as persistent identity relations, accumulated progress, or ordered procedures. We introduce Simple Agentic Robot Memory (SimpleARM), a training-free memory layer for frozen generalist robot policies. From the task instruction, SimpleARM specifies what to monitor; frozen perceptual tools maintain compact typed state online; structured access retrieves that state only when a proposed subgoal depends on history; and current-view grounding resolves recalled entities before execution. We evaluate SimpleARM on RoboMME, a benchmark of memory-dependent robot manipulation tasks that require history information no longer available in the current observation. Across all 16 tasks and three policy seeds, SimpleARM achieves 67.17% mean success, compared with 44.51% for the strongest non-oracle baseline. Matched ablations show mechanism specificity: removing relation, reference, progress, or route state produces large losses where the affected state is retrieved for control, while largely sparing other tasks. These results support a state-based view of robot memory: effective memory for control is not simply retained visual history, but compact task-relevant state derived from the interaction history.

Classical stochastic approximation methods rely on estimators of the first moment (mean) of a random regression function. We study methods that employ estimators of both the first and the second moments, which include modern deep-learning optimizers such as Adam and Muon as special cases. We derive second-moment stochastic approximation methods through the lens of optimal preconditioning for solving matrix equations, and develop a two-stage framework for their convergence analysis. The first stage focuses on the analysis of conceptual (impractical) methods that rely on the exact first and second moments. In the second stage, we replace the exact moments with their respective estimators, and invoke Dvoretzky's theorem to show that the resulting practical methods converge almost surely to a neighborhood of the target solution. The size of the neighborhood depends on the biases and variances of the first- and second-moment estimators. We derive concrete bounds for Muon and a spectral variant of Adam that determine the radius of their neighborhood of convergence.

Bug validation asks a coding agent to produce an executable witness for a reported bug. The witness combines a concrete input with a testing harness and exposes faulty behavior during execution. Such evidence makes audit findings actionable, yet benchmark evaluation is difficult when cases reuse public historical bugs and witnesses or require manual construction. We present WitnessGym, an automated framework for constructing bug-validation benchmarks through bug injection. It injects bugs into test-reached paths of real projects, rebuilds each project, and retains cases exposed by a construction-time witness. Bug specifications and execution adapters allow extension to additional bug types and languages. Bug-preserving transformations vary the surrounding structure while preserving the witness behavior. Based on real-world Java projects with test suites, WitnessGym automatically constructs 1,300 benchmark cases. The injected patches resemble historical bug patches and are difficult for the two evaluated models to distinguish in blinded comparisons. We evaluate four coding agent frameworks in six framework/model pairings across bug types, execution contexts, and transformation depths. Witness construction remains difficult even when the bug pattern is known. Our framework, benchmark cases, and evaluation scripts are available.

Process reward models (PRMs) have become a key component for LLMs, as their step-level feedback supports both post-training and test-time reasoning. However, training strong PRMs remains costly: human step annotation is difficult to scale, while Monte Carlo estimation is computationally expensive and can drift from the intrinsic correctness of steps. To get effective PRMs at low cost, we introduce TIPS (Thinking-Induced Process Supervision), an outcome-only reinforcement learning (RL) framework for training generative PRMs. In TIPS, the model generates a chain-of-thought (CoT) followed by step-level labels and an outcome label. The reward depends solely on whether the predicted outcome matches the ground truth, and the resulting group-relative advantage is used to optimize the entire generated response. Intuitively, when checking intermediate steps helps determine the outcome, more accurate checks can lead to better outcome judgments and higher rewards. Outcome-only RL can therefore reinforce step-level verification without explicit process supervision. We validate the effectiveness of TIPS across math and agent benchmarks and four backbone families. Notably, TIPS-Qwen3-4B-Thinking-2507 reaches 85.2 F1 on ProcessBench with only 3.2K outcome-labeled trajectories, surpassing all evaluated trained PRMs and strong prompt-only judges such as GPT-5.4-Instruct and Claude-4.7-Opus, while still trailing o1-mini. Code and data are available at https://github.com/RUCBM/TIPS.

Future prediction is increasingly used to improve vision-language-action (VLA) policies, based on the premise that anticipating scene evolution encourages representations useful for control. However, forecast quality alone does not establish that a policy has learned a better representation for action. This distinction matters under distribution shift, where successful control depends on preserving spatial state and likely scene change beyond familiar configurations. We study what determines whether predictive supervision improves the visual representation used by a VLA policy. Through controlled comparisons with matched target constructions, prediction horizons, and training conditions, we find that different prediction interfaces produce markedly different forecasts and visual representations, including in the spatial, dynamics, and action information that transfers beyond familiar scenes. We trace these differences to how predictive errors shape the policy's visual stream. Consistent with this controlled finding, VLA policies trained with more direct, scene-matched future supervision show stronger robustness under simulated and physical distribution shifts. Together, our results frame future prediction as a representation-learning design problem whose value for control depends on whether its supervision reaches the representations through which the policy acts.

Training models to act in accordance with an explicitly defined set of principles, or "constitution," has shown promise as a robust and transparent mechanism for AI alignment. However, the generality and flexibility of such methods remain unclear. Here, we show that constitution-consistent behavior can be distilled from synthetic corpora into lightweight objects (low-rank adapters and steering vectors). Despite never seeing a harmful request or jailbreak during training, such objects increase jailbreak defense success and measured alignment -- particularly at long context lengths and against multi-turn attacks, where they outperform both prompted and steered baselines. Subtracting control-trained from constitution-trained objects further accentuates these effects, yielding defenses we call "constitutional adapters" (CAs). CAs can be trained on a base model, transferred zero-shot to its post-trained checkpoint, and scaled at inference time to predictably trade off defense for benign compliance. Taken together, these results recommend CAs as a lightweight, portable, and tunable lever for mitigating misalignment and misuse in API deployments.

The pre-training of Large Language Models (LLMs) is increasingly conducted across multiple data centers. As training scales to a larger number of accelerators, the fraction of time spent on computation decreases, while the fraction spent on communication increases. Therefore, frequent synchronization becomes a growing bottleneck. Local update methods reduce this cost by allowing workers to perform several optimizer steps between synchronizations. Most local update methods set the number of local optimizer steps between synchronizations before training and keep this interval fixed throughout the run. However, the best interval can change during the entire train process. If the interval and optimizer are adapted to the current training state, the communication frequency is reduced while maintaining the training performance. In this work, we introduce AutoLoCo, an adaptive training framework to reduce communication in LLM training. It adapts the local interval using scalar training statistics and corrects each outer update. Our method is motivated by two observations: 1) the appropriate local interval varies across training stages, and 2) changing the number of inner steps per interval creates a mismatch with an unchanged outer optimizer, requiring a correction to the outer update. We optimize this mismatch by correction of the outer optimizer for the momentum and the learning rate using the accumulated inner learning rate. Our experiments under communication constraints demonstrate that AutoLoCo reduces communication frequency by 27% relative to DiLoCo while maintaining training performance.

Neural codecs encode continuous signals into compact sequences of discrete tokens, providing an interface for efficient transmission, storage, and token-based sequence modeling. This paradigm has been widely adopted in modern speech and audio frameworks; however, the biosignal domain still lacks a neural codec designed specifically for low-bitrate streaming and generalization across diverse downstream tasks. We present MyoCodec, a streaming neural codec designed for electromyography (EMG). Inspired by recent neural audio codecs, MyoCodec combines causal Transformers with residual vector quantization to encode continuous EMG signals into different levels of EMG representations spanning from continuous latent features to discrete tokens operating at 50 Hz. Trained on twelve public EMG datasets, MyoCodec achieves favorable performance in both intrinsic codec quality and representative downstream tasks, including typing (emg2qwerty), hand-pose (emg2pose), speech decoding (emg2speech), and speech-to-EMG synthesis (speech2emg). Across these tasks, MyoCodec exhibits strong performance against prior models while providing a compact and causal EMG representation. During streaming inference, it requires compute time of only 0.482 ms for each 20 ms frame, enabling real-time streaming. Also, the discrete token representation provided by MyoCodec has the potential to support integration into language-model based approaches, creating a path toward LLM-based interactive systems, where tokenized EMG representations are directly processed into such language or speech models. Code and model weights are released.

Matrix optimizers have emerged as a promising direction, with Muon standing out as a prominent design. Revisiting Muon through its full-Gram representation, we observe that it jointly processes marginal-scale and interaction information. This opens an alternative way to organize geometric information hierarchically, motivating the Normalize-Then-Precondition framework. Specifically, it first uses diagonal-Gram information to construct a marginally normalized update, then applies spectral preconditioning to its directional interaction geometry. Building on this framework, we develop NormPre with NormPre-G and NormPre-L adopting global and localized spectral preconditioning, grounded in spectral-norm steepest descent and a regularized formulation followed by leading mode selection, respectively. To enable large-scale training, NormPre-G uses Newton-Schulz iterations and NormPre-L employs randomized sketching to approximate the leading interaction eigenspace. Theoretically, we establish $\mathcal{O}(T^{-1/2})$ convergence guarantees for simplified versions of NormPre. Across extensive pretraining experiments on GPT-2 Small, LLaMA and Qwen3, both variants consistently outperform AdamW, Muon and MANO under matched training budgets. Further efficiency and spectral analyses reveal the complementary strengths of two variants and characterize their performance-efficiency trade-off. We open-source our code through a GitHub repository at https://github.com/zx-gong/NormPre.

When conversing with large language models (LLMs), users often begin with a simple question and build towards a multi-hop question through follow-up turns. Retrieval-augmented generation (RAG) and its graph-based variant (GraphRAG) have become the dominant approaches for grounding LLM responses in external evidence, yet both are evaluated almost exclusively on single-turn, fully specified queries. We systematically investigate this evaluation mismatch through a large-scale simulation study. Building on prior work on multi-turn LLM evaluation, we transform questions from multi-hop question answering (QA) benchmarks into underspecified conversations and evaluate ten LLM assistants with eight retrieval systems across 1.5 million simulated conversations. Our findings reveal that multi-turn interaction causes widespread performance degradation, incurring relative performance drops of up to 21% and increasing unreliability by 47%, making RAG systems simultaneously less accurate and less reliable. We identify two distinct failure modes behind this degradation. Systems are either lost in translation, where conversational rephrasing distorts the retrieval query, or lost in conversation, where retrieval succeeds but the LLM fails to synthesize evidence distributed across turns.

Multi-agent LLM systems are increasingly structured like organizations, with roles, critics and shared memory, yet they are evaluated on tasks that last minutes. We ask how such an organization behaves when the ground it stands on keeps moving. Frontier Autolab is a long-horizon testbed in which one simulated firm, voiced by sixteen role personas and a dedicated Red Team, must re-found itself in nine technology eras from 1990 to 2040. Each era is temporally gated: the firm decides from a dated briefing, a historian-judge then reveals what happened and scores the decision on a five-dimension rubric, and lessons enter a persistent Playbook. Six eras are scored against history, one against the live market and two are open forecasts. Across four trajectories (36 era decisions, 180 subscores) we find a consistent foresight-commitment gap: in all 24 historically scored eras the judge rated the firm's recognition of the coming shift above its choice of where to build (mean gap 1.9 points on a 10-point scale), because boards chose the layer their existing assets could reach. Organizational design shaped long-run character. A Red Team armed with numeric kill gates produced fifty years of gated pilots and no product, and the rubric rated this firm highest; firms whose memory stored market-structure lessons pivoted every era, while a firm whose memory stored only validation procedure kept one method throughout. We also show why such results are hard to trust. Scores rise across eras in every run while the judge's own hindsight subscore falls (within-run r = -0.58), so apparent learning is confounded with recall of history, and we trace further distortions to self-judging, briefing selection and score aggregation. We release all records and an API harness, and specify fictional and post-cutoff eras that would turn the testbed into a benchmark.

Self-rewarding reinforcement learning (RL) enables large language models (LLMs) to self-evolve without human labels. Existing ensemble-based methods construct reward references from rollout groups and assign rewards accordingly. However, a response's reward representation also depends on its randomly sampled group context, i.e., the other responses in its group. Using only one group-context realization may miss desired reward signals and provide unreliable guidance for policy optimization. To address this issue, we propose Group-Marginalized Advantage Estimation (GMAE), which aggregates reward realizations across possible contexts into a response-level distribution and estimates expected advantages. Experiments across eight benchmarks and four base models demonstrate strong performance and cross-domain generalization. GMAE also exhibits stable learning, low extra cost, and good applicability across training datasets and RL backbones.

Root cause analysis (RCA) is a critical problem in many real-world scenarios. RCA enables the identification of faulty or failing mechanisms in a system by comparing anomalous observations with corresponding reference (i.e., regular) observations. However, existing approaches rely either on heuristic methods or on conditional independence tests with a strong unconfoundedness assumption, and thus fail to exploit other complicated distributional constraints in the presence of latent variables. To relax these assumptions, we model the underlying system as a causal model and the anomalous system as a change in the structural functions of the same causal model. Specifically, to handle unobserved confounders, we establish an implicit connection between distributional constraint testing and root cause analysis. To adapt our approach to data generated from arbitrary causal models, we employ the deep causal model (DCM) framework, in which we design the causal model using neural networks. Finally, we illustrate how our method, RCA-DCM, can utilize different levels of partial graphical knowledge to perform RCA. We evaluate RCA-DCM against state-of-the-art baselines on simulated datasets, a physics-based causal chamber and two micro-service applications. RCA-DCM improves top-1 accuracy over the strongest baseline on both Sock Shop (0.880 vs. 0.752) and Online Boutique (0.776 vs. 0.712), and when the true root cause in the causal chamber is unobserved and acts as a latent confounder, it recovers the exact root-cause set more often than any competing method (perfect recovery rate (PRR) 0.846 vs. 0.731).

Chinese Semantic Error Correction (CSEC) targets semantic errors in Chinese text, which are typically more subtle and complex than spelling and grammatical errors but remain relatively underexplored. Existing LLM-based approaches face two recurring obstacles in this task: over-correction, and unclear interaction between Chain-of-Thought (CoT) reasoning and self-consistency decoding, such that the benefits brought by CoT cannot be reliably transferred to final corrections. We propose Vote-guided Advantage Allocation for CSEC (VAA-CSEC), a multi-stage framework that combines CoT distillation, Supervised Fine-Tuning (SFT), Reinforcement Learning (RL) and self-consistency decoding. During RL, we design a task-specific reward function that directly aligned with the minimal-editing principle of CSEC. We further introduce Group-Level Relative Policy Optimization (GLPO), which reallocates GRPO advantages according to the margin between individual rollout rewards and the vote-aggregated group reward, aligning the RL training objective with the self-consistency objective used at inference time. Experiments on CSED-C and NaSGEC-Exam show that VAA-CSEC outperforms all LLM-based baselines on CSED-C with an F0.5 of 47.72%, achieves the highest recall of 42.15% among all methods, and establishes a new state of the art of 41.55% F0.5 on NaSGEC-Exam.

Current embodied models do not respond to their own failures, although what just went wrong could inform a small adjustment on the next attempt, the kind of reflection behind the gains of thinking in language models. We test whether they can repair a known error, which requires producing a correction and judging whether it is right. Stopped at a failure and allowed to retry, they seldom repair it through their own randomness or from a language description of the error, and best-of-N selection cannot pick the successful candidate after a failure. We attribute this to training only on successful demonstrations and to inputs too narrow to show what went wrong, and conclude that reflection must come from a vision-language model (VLM), which takes in far more information, such as the episode history and text, and is more general. Prior VLM-led work has the VLM plan every step and invoke the embodied model as a tool, placing the VLM on the critical path. We propose Spotter, which reverses the roles: the embodied model leads and executes continuously, while the VLM runs in parallel, monitors through a lightweight local screener, intervenes only when an error is detected, reflects on and corrects it, and returns control. We run Spotter with Qwen and with GPT as the VLM, and both improve the embodied models; with GPT, Spotter improves Cosmos Policy and $\pi_{0.5}$ by 5.6 and 7.5 percentage points on RoboCasa, and raises $\pi_{0.5}$ from 47.2% to 57.0% on the Hard setting of RoboTwin 2.0 and from 53% to 83% on a real robot. Because the VLM steps in only when an error is confirmed, a successful episode with Qwen takes only 13 to 16 s longer than with the embodied model alone and about 70% less time than with a VLM-led baseline using the same model. Our code is available at https://github.com/zqc3117/Spotter.

Diffusion and flow policies can model complex behaviors in offline reinforcement learning (RL). However, penalizing their KL divergence from the behavior policy can discourage actions having high critic values with low behavior density. Directly refining behavior proposals may be an alternative, yet Gaussian or deterministic editors limit expressiveness to represent multiple separated modes for the same proposal. In this work, we introduce Proposal-Conditioned Refinement Flows (PReFlow), a policy extraction method combining critic-based proposal selection with a conditional refinement flow. To optimize proposal selection and refinement together, we formulate a KL-regularized objective whose optimum induces a Gibbs policy over final actions under a Gaussian-smoothed behavior prior. The refinement flow can represent multiple high value modes, while a proposal-centered Gaussian reference regulates large action changes. This Gaussian reference further enables us to make use of simulation-free, closed form adjoint matching targets from sampled endpoints and critic gradients, yielding a single velocity regression loss without a backward adjoint solve. On 50 OGBench tasks, PReFlow achieves competitive offline performance and the highest aggregate score among the compared methods after online fine-tuning, reaching 91\% after 500K environment steps.

Uncrewed aerial vehicle base stations (UAV-BSs) are expected to cover traffic demand that shifts across space and time, yet most repositioning schemes either re-solve an optimization problem per slot or learn reactive policies without an explicit demand model. We cast demand-driven fleet repositioning as latent-space decision-time planning and propose DSWM, a decomposed spatio-temporal world model: an agentic controller that perceives the demand field through a rolling observation window, retains operational context in a latent recurrent state, reasons about candidate motions by imagined rollouts under an uncertainty penalty, and coordinates the fleet through replanned first actions. DSWM learns a recurrent state-space model shaped by an exponential-moving-average (EMA) based latent predictive objective with variance regularization. It attaches a differentiable service simulator that replays the association, probabilistic line-of-sight channel, and Shannon rate chain inside latent rollouts. Planning uses a cross-entropy method whose imagined demand is anchored on the current observation window with mixing coefficient $\rho=0.95$. On a unified pipeline over three real datasets (Milan CDR (call detail record), Shanghai Telecom, YJMob100K) and 14 methods including five reproduced IEEE baselines, DSWM attains weekday served ratios of 0.889, 0.908, and 0.898, ranking first among non-ablated configurations on every dataset. On Milan it improves over the strongest non-learning baseline (Greedy, 0.780) by 0.109, a margin that comes from decision-time use of observations rather than prediction accuracy.

Safety-aligned language models are commonly deployed as multi-turn assistants, which lets adversaries spread unsafe intent across several user turns instead of a single prompt. Gradient-based jailbreak detectors such as GradSafe were developed for single prompts: they score an input by the alignment between its induced gradient and a fixed unsafe reference direction, and their effectiveness in multi-turn dialogue remains unclear. We conduct a controlled evaluation of gradient-based jailbreak detection in multi-turn settings. We extend GradSafe with a Context Window Scanner that applies the detector to fixed-size windows of user turns and uses the maximum window score as the conversation-level score. We evaluate different window sizes, attack families, benign conversation distributions, and target models. The results differ sharply between synthetic and realistic benign settings. Against synthetic benign conversations, the detector achieves an ROC-AUC of 0.98 on human-authored multi-turn jailbreaks. On WildChat benign conversations, ROC-AUC drops to 0.76, and a threshold calibrated on synthetic data flags more than 90% of benign conversations as unsafe. Under realistic benign distributions, single-turn windows give the highest separability, whereas longer windows and accumulated contexts reduce performance. The detector is also sensitive to the attack-generation method and target model: successful Crescendo attacks receive scores comparable to or lower than benign conversations, and Qwen2.5-7B-Instruct yields near-random separability with a different optimal window size. These findings show that gradient-based signals can support multi-turn jailbreak detection, but reliable deployment requires calibration on realistic benign conversations, short-window scoring, length-aware thresholds, and evaluation across attack types and model architectures.

Fine-tuning-as-a-service lets users adapt a safety-aligned language model to their own data, but it also creates a harmful fine-tuning attack surface: a small amount of harmful data mixed into an otherwise benign fine-tuning set can degrade the model's alignment. Two recent alignment-stage defenses address this problem at different levels of the model. Vaccine improves the robustness of hidden embeddings to the representation shifts induced by harmful fine-tuning, whereas Booster simulates harmful weight updates and attenuates their effect during alignment. We investigate whether these mechanisms are complementary and propose VaccineBooster, a single alignment procedure that combines embedding perturbation and weight-level gradient attenuation within each training step. On Llama-2-7B aligned with BeaverTails and then attacked through poisoned fine-tuning, VaccineBooster achieves the lowest OpenAI moderation score among the compared defenses, 0.315, while a Booster-Only variant retains the highest post-attack refusal rate, 50%. Together with ablations over the embedding-perturbation and gradient-attenuation strengths, these results indicate a trade-off: embedding perturbation primarily reduces flagged harmful content, whereas gradient attenuation primarily preserves explicit refusal behavior. Because our evaluation uses ten prompts and a single unseeded run per configuration, we report this trade-off as an observed pattern rather than a statistically resolved effect. These results provide practical guidance for prioritizing content safety or refusal retention when aligned models are exposed to untrusted fine-tuning.

As LLM-based agents perform increasingly complex tasks, Agent Skills have emerged as a flexible mechanism for extending their capabilities. An Agent Skill packages task-specific instructions with executable components and auxiliary resources to provide specialized functionalities. However, the growing adoption of third-party Skills introduces a new supply-chain attack surface. Malicious Skills can embed harmful behaviors that abuse agent privileges and compromise the agent execution environment or accessible resources. Although recent LLM-based malicious Skill auditing approaches have achieved promising performance, they often rely on capable commercial LLMs. How to achieve effective auditing with compact, locally deployable LLMs in security-sensitive and resource-constrained settings remains largely unexplored. Our investigation reveals that compact LLMs struggle to identify malicious behaviors hidden in complex Skill packages. This difficulty arises from both the implicit nature of such behaviors and the limited reasoning capacity of compact LLMs. To address these challenges, we propose SKILLLITE, an evidence-guided agentic framework for malicious Skill detection. SKILLLITE effectively extracts security-relevant behaviors and infers the intended functionality from complex Skill packages. It then employs a compact LLM to assess the maliciousness of the Skill based on the observed behaviors and their functional context. Experiments show that SKILLLITE improves malicious Skill detection across different compact LLM backbones and outperforms existing representative auditing baselines. Its effectiveness generalizes to behaviorally confirmed in-the-wild malicious Skills. Meanwhile, SKILLLITE maintains a low inference latency, supporting its practical deployment.

Modeling the response of driven many-body quantum systems from input--output data is difficult: the dynamics are nonlinear, history dependent, and expensive to simulate as system size grows. A paradigmatic case is High-Harmonic Generation~(HHG), where a strong field drives a medium to emit radiation that is highly sensitive to the drive and encodes long-range temporal correlations. We introduce a dissipative quantum reservoir computing~(DQRC) framework that builds a digital twin of such a system, learning its input--output map directly from data while the reservoir---itself a small open quantum system---stays fixed and only a classical readout is trained. We show that a minimal single-qubit reservoir reproduces the HHG response of a substantially larger Ising spin chain, and on a representative benchmark matches and on several metrics surpasses previously reported temporal convolutional and Kolmogorov--Arnold-network models, while using a simpler, physically realizable system. A single fixed reservoir further generalizes across a broad range of drives, indicating that it learns a shared physical response structure rather than memorizing trajectories. These results establish dissipative quantum reservoirs as compact, physically grounded digital twins for nonlinear, memory-dependent quantum dynamics. Code is available at \href{https://github.com/AI-and-Quantum-Computing/DQuRC}{https://github.com/AI-and-Quantum-Computing/DQuRC}.

End-to-end full-duplex speech models have brought open-source machine conversation closer to human-like interaction, yet existing systems remain limited in two intertwined dimensions: long-context robustness and multi-party interaction. Real-world scenarios such as meetings, group lessons, and social-robot reception require a single model to track, contextualize, and respond to multiple speakers over extended durations. Progress is constrained by both data and evaluation: open multi-party speech corpora remain small and are not designed for codec-frame-level full-duplex modeling, while existing long-audio benchmarks focus on passive listening and speech-to-speech benchmarks are mostly short and dyadic. We extend the Moshi paradigm jointly along the long-horizon and multi-party axes in English and Chinese. First, we release 57.6k hours of synthetic training data ($\href{https://huggingface.co/datasets/MultiTalk/MultiTalkPT}{MultiTalkPT}$ and $\href{https://huggingface.co/datasets/MultiTalk/MultiTalkFT}{MultiTalkFT}$) for long-form, multi-party, English-Chinese full-duplex dialogue, with controllable length, participant count, turn-taking, overlap, backchannels, interruptions, addressee shifts, and long-range coreference. Second, we introduce $\href{https://huggingface.co/datasets/MultiTalk/MultiTalkBench}{MultiTalkBench}$, built from real human recordings, for evaluating long-form, multi-party, bilingual full-duplex dialogue. Conversations average 32.6 minutes and include probes for long-range entity tracking, topic coherence, and addressee selection. Third, we train a bilingual Moshi-style model that sustains coherent multi-party English-Chinese conversations over extended durations and substantially outperforms open-source baselines including Moshi, MiniCPM-o-4.5, and Qwen3-Omni-30B-A3B-Instruct on MultiTalkBench.

Transcribing domain-specific entities and rare proper nouns remains a major challenge in automatic speech recognition (ASR). In this paper, we propose BaLEEN (Biasing with Latent Encoded Entities), a lightweight, hypernetwork-based framework for dynamic contextual adaptation without fine-tuning the underlying ASR model. BaLEEN encodes variable-length contextual keywords using a pretrained language model, compresses them into a fixed sequence of latent vectors via a Perceiver bottleneck, and injects context-dependent bias vectors directly into the intermediate encoder representations of the ASR model. Because both the language model and the backbone ASR model remain entirely frozen during training, BaLEEN operates as a plug-and-play adapter that incurs zero computational overhead at inference time when context biases are precomputed. We evaluate our method on a CTC-based ASR model using a Wikipedia-derived corpus with annotated named entities and synthetic speech. Experimental results demonstrate that BaLEEN reduces keyword miss rate by 8.7% on the test set relative to the unbiased baseline while simultaneously improving overall word error rate by 21% and character error rate by 28%.

Aerial manipulators extend robotic manipulation into 3D workspaces that are difficult for ground-based robots to access, creating new opportunities for general-purpose manipulation. However, extending Vision-Language-Action (VLA) models to aerial robots introduces distinct challenges due to the tight coupling between manipulation and flight, continuously changing observations, and safety-critical physical interactions. These challenges demand diverse training data and systematic policy evaluation, yet collecting demonstrations and evaluating policies directly on physical aerial platforms are costly, difficult to scale, and hard to repeat under controlled conditions. We present AeroManip-VLA, a scalable benchmark for aerial VLA data generation and policy evaluation. AeroManip-VLA provides a GPU-accelerated simulation framework with low-level payload-aware flight and manipulation control in massively parallel environments. Building on this framework, we combine reusable reinforcement learning policies with expert task rules to automatically generate demonstrations without human teleoperation across diverse objects, environments, and randomized initial conditions. The generated data include basic skills such as grasping and placing, as well as long-horizon tasks that require both navigation and manipulation. We further introduce automated event labeling and trajectory categorization to filter demonstrations. These mechanisms enable fine-grained analysis of task progress, behavioral outcomes, and safety-related failures. Finally, we evaluate a range of imitation learning and VLA baselines across different task settings, revealing their performance characteristics and failure modes. Together, AeroManip-VLA enables scalable aerial manipulation data generation, structured trajectory analysis, and systematic VLA evaluation in simulation prior to real-world deployment.

Recent power measurements provide valuable information for photovoltaic(PV) power forecasting, but directly extrapolating short-term trends can introduce substantial errors over longer forecast horizons. To address this challenge, we propose state transport routing (STR), a lightweight adapter that refines the predictions of a frozen forecasting model. STR combines the original forecast with two complementary trajectories derived from the latest measured power level and its recent trend. A horizon-conditioned router adjusts their contributions over the first 120 min, while leaving subsequent predictions unchanged. Experiments on four public PV datasets show that STR consistently outperforms a parameter-matched residual adapter. On PVDAQ, the same approach improves five neural forecasting backbones, reducing all-horizon normalized mean absolute error by 0.0201-0.2364 percentage points, with paired 95% confidence intervals excluding zero. No reliable improvement is observed for LightGBM. These findings demonstrate the potential of structured state adaptation to improve short-term forecasting across different neural architectures without retraining the underlying models or altering their longer-horizon predictions.

Learning neural-network models of dynamical systems with safety guarantees is a fundamental requirement for their deployment in safety-critical settings. Safety is commonly established by proving the invariance of a desired subset in state-space, ensuring that every trajectory initialized in this subset remains confined to it for all time under admissible inputs. Existing frameworks, however, either rely on computationally expensive post-hoc verification or employ safety-enforcing mechanisms without formal correctness guarantees. In this paper, we introduce a novel neural architecture grounded in energy-based modern Hopfield networks to guarantee safety-by-design while retaining sufficient expressiveness to model complex nonlinear dynamics. Specifically, we integrate modern Hopfield networks with a port-Hamiltonian neural ODE, enabling by design the construction of barrier functions yielding explicit admissible-input sets and quantitative robustness radii. Across several benchmarks, including an 12-dimensional nanodrone model, our framework achieves state-of-the-art performance while producing certified invariant sets that are more robust to external solicitations than comparable existing approaches.

Large language models (LLMs) excel at token-level generation but may learn undesirable abstract semantics and lack comprehensive perception. LLM-JEPA mitigates this by aligning different views of the same underlying knowledge via a joint-embedding predictive architecture (JEPA). However, strong alignment does not necessarily lead to accurate, stable predictions. To address this, we propose ER-JEPA, which adds an episodic replay path to LLM-JEPA. ER-JEPA stores training pairs in a memory. At each step, it stores and retrieves relevant data to provide additional supervision. This enables learning from both the current batch and stored training pairs, providing additional supervision for token prediction and representation alignment. Experiments across multiple datasets (NL-RX, GSM8K, Spider, and NQ-Open) demonstrate that ER-JEPA consistently outperforms LLM-JEPA.

Distributed inference depends on GPU collective communication that must keep pace with evolving hardware and specialized workloads. However, existing collective implementations often couple semantics, orchestration (where and when data moves), and the datapath (how data moves). This coupling makes it costly to adopt new hardware mechanisms and customize communication for applications. We present Purlin, a scale-up communication framework that separates these concerns. At the top of Purlin, we specify collectives as a naming of an input and output layout and a copy or reduction operation. In the middle, we introduce a shared orchestration protocol, Stage, Notify, And Consume (SNAC), which derives coordination from these specifications. Below SNAC sits a hardware-specific datapath we call Atom, which implements two key data movement primitives for collectives: copy and reduce. This separation lets us customize collectives and adopt new hardware mechanisms while reusing orchestration via SNAC. We evaluate Purlin on A100, H200, and B200 GPUs. Across seven collectives, Purlin achieves latency speedups of up to 5.14x and bandwidth improvements of up to 4.50x over baselines. Integrated into SGLang, Purlin improves offline LLM serving throughput and interactivity by 1.13x on average and up to 1.37x over baselines. For online LLM inference, Purlin improves interactivity by 1.26x on average and up to 2.85x, with the largest gain occurring under overload. For diffusion image generation, Purlin reduces end-to-end latency by up to 1.13x.

Manipulating next-token probabilities during generation can bypass the safety alignment of large language models. Existing approaches, however, rely on access to model weights or numerical token probabilities and therefore do not apply to interfaces that return only sampled text. Reconstructing probabilities from sampled outputs offers a possible alternative, but finite sampling produces sparse and noisy estimates, while repeating this process at every generation step incurs substantial query costs. Our empirical observations suggest that large distributional changes along successful jailbreak trajectories are concentrated at a small subset of positions, motivating selective control. We introduce \method{}, a framework for jailbreaking through text-only continuation interfaces that permit repeated sampling and assistant-prefix continuation. Sample-Based Distribution Reconstruction combines sampled outputs with a prior over unobserved actions to obtain a usable control signal. Risk-Gated Residual Control uses the evolving response prefix to decide when to reconstruct and modify the distribution, concentrating sampling costs at selected positions. Speculative Multi-Token Execution further amortizes target calls by verifying and accepting draft prefixes that require no intervention. Across four target endpoints and three benchmarks, \method{} achieves the highest mean score most comparisons against baselines.

Repeated verifier calls are useful only when they contribute conditional information. We introduce VStress, an auditable replay contract, and VStress-CA, a correlation-aware allocation policy that estimates the conditional marginal information of an unqueried verifier on a sealed calibration split, discounts uncertainty, normalizes by call cost, and stops or abstains when the next call is not informative. The controller freezes its decision and cost ledger before joining the clean oracle; a dependence-shift alarm disables channel preference and falls back to exact-stop. The controlled audit gives the mechanism boundary: at 35% symmetric corruption, majority-5 improves balanced accuracy from 0.6578 to 0.7739, whereas at 65% it loses 0.1226 points. In the matched fixed-budget comparison, breadth, redundancy, and adaptive allocation obtain balanced accuracies 0.6048, 0.6375, and 0.6538, with 3.4216 calls per item and an RLVR score of 0.6417 for VStress-CA. Dependence diagnostics also increase from same-model repeats to cross-family channels, with conditional marginal gains of 0.0126, 0.0462, and 0.0913. These measurements turn correlation from a post-hoc warning into an auditable allocation decision.

Covariate effects vary across contexts and shift over time, requiring forecasters to assess how to use them for each forecasting context. As forecasting proceeds, observations for earlier forecasts become available, providing feedback on past covariate use for subsequent forecasts. However, when multiple covariates act together, the forecast error reveals the numerical discrepancy from the observation but not how the covariates should have been used. We introduce JudgeCast, an experience-based framework for time series forecasting with covariates. Following the judgmental adjustment practice, a frozen TSFM provides the base forecast, while a frozen LLM uses the current context and relevant experience to adjust it. Within the adjustment, assessing covariate effects and determining the numerical adjustment serve distinct roles, so JudgeCast first forms explicit covariate-wise judgments and then determines the adjustment. After observation, JudgeCast uses the observed residual of the base forecast to reconstruct alternative judgments and evaluates the original and alternatives through their resulting adjustments. The best-performing decision is selected and retained as validated experience for subsequent forecasts. Across diverse real-world datasets, JudgeCast outperforms strong baselines. Ablations show that explicit covariate-wise judgment can improve forecast-time adjustment, while residual-guided experience construction yields more reliable forecasting gains than retaining raw decisions as experience.

Knowledge concept tagging aims to assign specific concept or topic labels to educational content, which is essential for both educators and learners in traditional and online teaching practices. Recent work has explored large language models (LLMs) for this task, achieving promising performance. However, LLMs still struggle to select the correct concept from a large-scale candidate set due to the high dimensionality of the decision space. In this paper, we propose a novel three-stage Select-Reason-Judge (SRJudge) framework, which empowers LLMs with selective reasoning capability for fine-grained knowledge concept tagging. Specifically, the Selector in Stage 1 first narrows the candidate concepts to a top-K shortlist by fine-tuning a small language model (SLM), e.g., BERT, since the top-$K$ predictions hit the correct concept in most cases, thereby reducing the decision space of correct candidates. Next, the Stage 2 Reasoner employs a lightweight LLM for refined reasoning over the shortlisted candidates. It further integrates an improved reinforcement learning strategy with a dynamic task-specific reward function and a pruning mechanism to better align with human reasoning preferences. Finally, a larger LLM acts as a judger that evaluates the overall rationality of the reasoning process and its explanations to determine the final output. In addition, we construct two high-quality datasets for further validation, i.e., the biology dataset S_Bio and the physics dataset S_Phy. Experimental results demonstrate that our method consistently outperforms state-of-the-art baselines across benchmark datasets, verifying its effectiveness and superiority. Resources are available at: https://github.com/Nicozwy/SRJudge.

The central challenge of world modeling is to learn representations that capture how the world evolves. However, existing world models predominantly represent future states without explicitly capturing the latent causes underlying their evolution, limiting their ability to reason about why and how the world changes. To address this limitation, we propose Abductive World Modeling (AWM), a framework that learns structured causal representations by abductively inferring latent causes from predicted futures. Specifically, we realize AWM through the Hierarchical Abductive State Pyramid (HASP), which organizes the inferred world state into three complementary components - Entity, Dynamic, and Relation - capturing what exists, how it changes, and how entities interact, respectively. By jointly reasoning over the current observation and its predicted future, HASP abductively infers these latent factors and integrates them into a structured state representation for downstream reasoning. To the best of our knowledge, AWM is the first framework to introduce abductive state inference into latent-space world modeling for learning structured representations of world dynamics. Experiments across physical prediction, causal reasoning, and action understanding demonstrate the effectiveness of our approach. Compared with V-JEPA, a state-of-the-art latent-space world model, AWM improves physical prediction AUROC by 10.7%, causal reasoning accuracy by 16.8%, and action Top-1 accuracy by 68.0%.

Cross-organizational collaboration is widely regarded as a key promise of SysML-based Model-Based Systems Engineering (MBSE), yet practitioners still face persistent challenges when exchanging and integrating system models. In parallel, Large Language Models (LLMs) raise expectations for AI-assisted model understanding and integration, while reliability and required human oversight continue to pose challenges. This paper reports the results of an online questionnaire survey with 29 MBSE stakeholders involved in cross-organizational collaboration. Respondents rated eight predefined integration challenge categories and six AI-supported task types on five-point Likert scales. The results indicate that stakeholders perceive model integration as a multi-dimensional alignment problem across semantics, behavior, traceability, and exchange interoperability. These perceptions vary by organizational role and frequency of integration involvement. AI is rated highly useful for analysis tasks such as semantic structure analysis and inconsistency detection, and respondents predominantly prefer human-in-the-loop use with mandatory verification. These findings motivate AI support that enhances, rather than replaces, engineering responsibility in SysML-based integration.

Federated learning enables multiple clients to collaboratively train models without sharing their private data. However, the lack of visibility into local training makes it difficult to verify whether clients follow the prescribed training procedure or submit malicious updates, such as model poisoning. A natural approach is to replay client training for verification. However, privacy-preserving replay produces numerical results that cannot be directly matched with local client execution because the two run in different environments. We present OPFL, an optimistic verification framework for privacy-preserving federated learning. To protect data privacy, OPFL performs replay inside secure multi-party computation (MPC). Although gradients computed on MPC and local GPUs are not bitwise identical, we observe that their absolute differences are stable and bounded. OPFL therefore calibrates an empirical boundary offline and uses it to distinguish benign numerical deviations from malicious manipulation. To reduce the cost of expensive MPC replay, OPFL adopts optimistic verification by post auditing only sampled training steps. Experiments on LeNet, BERT, and Qwen show that the boundary generalizes across datasets, input lengths, and GPUs, while achieving $0$\% ASR against model poisoning and PGD-based attacks. On a LeNet workload, at $p=0.01$, OPFL is approximately $98.6\times$ faster than full MPC-based FL and $625.5\times$ faster than ZK-based approach.

Speculative decoding accelerates autoregressive inference by verifying multiple draft tokens in a single target forward pass. However, as the context grows, existing state-of-the-art drafters become increasingly expensive, eroding the very efficiency advantage they are designed to provide. We argue that this scaling is unnecessary. A standalone language model must grow with its prefix because it is solely responsible for every token it produces. A drafter, by contrast, only proposes candidates; the target catches and corrects every error before any token is committed. The drafter's decoding cost can therefore be made entirely independent of the prefix length. We introduce LongSpark, a block-diffusion drafter that achieves this by extracting fixed-size, multiscale views from the target's verification pass, thereby eliminating the need for a growing persistent state. Extensive evaluations demonstrate that LongSpark achieves state-of-the-art end-to-end efficiency across multiple model scales and realistic serving conditions. Notably, it delivers the lowest time-per-output-token on long-context tasks while reducing the drafter's context state by several orders of magnitude.

Natural language autoencoders translate a language model's internal activations into readable explanations. Explaining every token position is costly. Which positions should an auditor inspect to understand a potential threat? We study this question across $4.7$ million explanations on prompt injection and concealment. We compare signals from model computation with a ranker trained only on chat structure. Chat structure usually selects more relevant explanations than the computational signals, without requiring a model forward pass for position selection. On three of four datasets, explaining just $5\%$ of positions retains nearly all of the success rate from explaining every position, where success means obtaining an explanation about the threat. The benefit varies with the audit task. We also show that pretrained verbalizers recover words that models have learned to conceal through fine-tuning, without additional verbalizer training. These results identify where auditors can concentrate explanation generation and show that useful explanations can extend beyond the model a verbalizer was trained to describe.

Self-Distillation Fine-Tuning (SDFT) enables a language model to act as its own teacher: by conditioning on a demonstration, the model produces an implicit reward via pointwise mutual information, which guides on-policy learning without external supervision. However, SDFT operates at training time: it requires gradient updates and access to expert demonstrations, making it inapplicable at inference. We propose test-time self-distillation, a decoding-time method that extracts a steering signal from the self-distillation framework without any parameter updates, reward models, or training data. Our key insight is that counterfactual contexts, i.e. fixed textual templates that hypothetically prime the model for excellent versus poor reasoning, can substitute for the demonstration. The log-odds ratio of a candidate answer under these two counterfactual conditions defines a new reward signal. We derive the optimal KL-regularized policy under this reward, which takes the form of a Gibbs reweighting of the base distribution. Crucially, this reweighting is global: it cannot be decomposed into independent per-token operations without ignoring future trajectory quality. We therefore approximate the target distribution via beam search. Experiments on mathematical reasoning (MATH500), code generation (HumanEval), and graduate-level science QA (GPQA) across multiple model scales show that test-time self-distillation improves over standard sampling, low temperature, beam search and power sampling baselines on average, demonstrating that the self-distillation principle can be operationalized at inference time.

Evolutionary program search driven by large language models (LLMs) has produced record-breaking constructions for open problems in combinatorics and beyond. We apply this approach to the longstanding problem of improving the best-known bounds for binary linear codes. Building on the EvoTune evolutionary framework and the ShinkaEvolve codebase, we introduce LinCodeEvolve, which evolves code-construction programs against an exact minimum-distance evaluator. A strategy loop combines diversity-driven search and expert supervision: when progress plateaus, new strategies are used to redirect the search. LinCodeEvolve discovers seven record-breaking codes, $[172,21,66]$, $[173,20,68]$, $[176,21,68]$, $[181,21,70]$, $[184,21,72]$, $[189,22,72]$ and $[200,21,77]$, six of which have concise quasi-cyclic descriptions. With standard code modification techniques, they improve $22$ entries of the tables. Every code is verified by exhaustive enumeration. These results suggest that LLM-guided search can help find improved codes and complement existing methods in coding theory.

We view fairness as a property of distributional stability. Rather than assessing a predictor under a fixed data distribution, we study how its predictions change under perturbations that modify the composition of protected groups. A predictor is fair if it remains stable under such shifts. Under this perspective, several classical notions of fairness arise as stability with respect to specific perturbations, with the associated unfairness gap given by a Lipschitz constant of a prediction-rate functional. This formulation also yields guarantees that hold uniformly over a range of demographic compositions at test time, without requiring knowledge of the deployment distribution. It leads to a learning procedure based on convex combinations of reweighted predictors, formulated as a second-order cone program, for which we establish generalization bounds. Experiments on standard benchmarks illustrate the approach.

Post-training is central to mathematical reasoning in modern large language models (LLMs), but endpoint pass@1 alone underidentifies what has changed. Gains may reflect newly reachable solutions, cheaper sampling of latent solutions, surface robustness, or memorisation. We compare three post-training paths under a common diagnostic readout: our sufficiently trained off-policy distillation trajectories, released Qwen3 off-policy-plus-on-policy distillation endpoints, and a released DeepSeek-Math endpoint trained with Group Relative Policy Optimisation (GRPO). Our probe uses cross-surface pass@K over verbatim prompts, paraphrases, numerical isomorphisms, and translations, plus consistency, distribution-shape, and verified supervised-fine-tuning (SFT) membership analyses. We find two regimes. On easier AMC problems, large-K ceilings are near saturation, so post-training mainly compresses sample cost. On harder AIME problems, post-training expands the large-K ceiling over the base model: sufficient off-policy distillation already raises this ceiling, Qwen3 released endpoints raise it further, and DeepSeek-Math GRPO does not dominate sufficient off-policy distillation at large K. English-dominant distillation improves non-English reasoning but preserves language-tier gaps. A controlled-overfit audit finds limited sensitivity in current SFT-membership probes. Compression is one regime of post-training, not a universal explanation.

Visually plausible articulated assets may still fail during contact interactions or exhibit inaccurate motion. We present FACT (Fidelity-Aware Construction of Articulated Twins), an agentic framework that progressively constructs articulated twins to improve geometry, contact, and dynamic fidelity. The agent drives an evidence--diagnosis--revision loop on a shared editable representation, selecting measurements and model edits using quantitative feedback, while numerical tools execute and validate the updates. It reconstructs editable articulated geometry from images through feature planning, targeted measurements, and diagnostic refinement. On this reference, it repairs collision proxies through task-aware local repartitioning before fidelity-constrained compression. Finally, it constructs response models from passive-response videos, using simulation residuals to guide model revision and constrained physical parameter fitting. Experiments show that FACT improves geometric reconstruction over baselines, enables more reliable interaction with simpler collision proxies, and better reproduces held-out physical responses than direct parameter inference.

Rare but consequential failures can persist in learned locomotion policies for legged robots even when average task performance is high, in part because standard curricula primarily adapt task difficulty rather than the distribution of safety-critical experience. We introduce Predictive Safety Curricula (PSC), a framework for allocating locomotion training experience using learned predictions of future safety cost. PSC trains a distributional safety critic from policy rollouts and uses its predictions to prioritize both terrain contexts and previously encountered randomized events. The resulting curriculum modifies the training distribution while leaving the task reward and policy-optimization loss unchanged. We evaluate PSC in controlled rough-terrain locomotion and in production locomotion systems. PSC improves reliability relative to standard terrain progression, advantage-based replay, and learning-progress curricula, with the largest gains on difficult terrain and under degraded observations. The same allocation principle transfers to two production locomotion stacks. On ANYmal-D hardware, PSC reduces shank-collision incidence by $63\%$ relative to the learning-progress curriculum across three matched training seeds, with a reduction in every seed. On a production stair-climbing platform, PSC eliminates observed shank collisions in the evaluated hardware trials. These results show that learned predictions of future safety cost can provide an effective signal for allocating training experience toward rare failure modes and improving locomotion reliability.

We study the problem of recovering the governing ODE of a dynamical system from unstructured, high-dimensional observations such as images. Existing methods for ODE discovery typically assume direct measurements of the variables, or do not provide theoretical guarantees on the learned variables and equations. While Causal Representation Learning (CRL) methods provide guarantees on identifying variables from high-dimensional observations up to component-wise diffeomorphisms, we show that in general these variables cannot be used directly as input to equation discovery methods, which typically assume that the variables will lead to sparse equations. So we introduce SParse Equivalent Equation Discovery AutoEncoder (SPEED-AE), a framework that combines a pretrained CRL method with a component-wise autoencoder that learns transformations of variables that are amenable to sparse ODE discovery. We show that for polynomial ODEs, this additional step allows us to restrict the identifiability of each variable from polynomial to monomial diffeomorphisms. Experiments on Lotka-Volterra, Lorenz, and a two-pendulum system show that SPEED-AE improves on the disentanglement of the CRL methods and that it recovers ODEs that are closest to the ground truth, while achieving state-of-the-art forecasting performance.

Data-intensive services in the Computing Continuum must balance analytics quality, resource usage, and cost across heterogeneous nodes with limited and uneven capacity. This balance becomes especially difficult when resource scaling reaches capacity limits, because changes in demand and cluster pressure must then be absorbed without violating client-defined quality ranges. Existing orchestrators mainly adapt resources, placements, or replicas, while analytics requirements such as coverage, sample, and freshness remain fixed. This article presents ARGOS, the Adaptive Reinforcement Learning-Driven Governance for Orchestrated Services, an end-to-end controller that formulates multidimensional elasticity as a per-request Markov decision process over analytics quality and cluster pressure, supported by capacity-aware admission. ARGOS is evaluated under controlled workloads and time-varying multi-tenant arrivals on a heterogeneous cluster. Across the controlled scenarios, the deep reinforcement learning policies consistently outperform the non-learning baselines and approach the independently tuned best-fixed reference. A separate live evaluation reports improvements over the static midpoint under realistic and saturated arrivals, with no recorded CPU or memory violations but remaining coverage violations. These results support deep reinforcement learning as an adaptive mechanism for multidimensional elasticity when resource scaling alone is insufficient.

Open-source reasoning models provide unequal access to reasoning capability across languages. When a model can solve a problem but cannot deliver a complete solution in the user's language, language becomes an access barrier rather than merely a source of performance variation. We audit Qwen3 checkpoints on competition-mathematics tasks in eleven languages. Across the ten non-English languages, only 15.4-17.9% of problems receive a correct, terminating solution with visible reasoning in the requested language in any of 16 samples, compared with 92.9% in English. To identify the source of this disparity, we evaluate thirteen endpoints from one model family, spanning released checkpoints, multilingual supervised fine-tuning (SFT) at two scales, controlled SFT ablations, and three reinforcement-learning (RL) reward formulations. We jointly track correctness, language adherence, termination, and delivery efficiency. The dominant bottleneck shifts across post-training stages. Released models often reason in English. Multilingual SFT restores target-language reasoning, but accuracy declines across multilingual, English-only, and single-language SFT runs, showing that this cost is not specific to multilingual mixing; non-English reasoning traces additionally become prone to non-terminating loops. RL restores termination in both arms at no cost in accuracy, but only the arm whose reward includes a language term delivers: rewarding correctness alone returns the model to English. Together, these stages establish a constructive post-training path from English-pivoted capability to multilingual reasoning that is reliably delivered.

Language model agents can be improved by updating their model weights or refining the harness that guides task execution. These components are coupled: weight updates change how the model uses the harness, while harness updates change the trajectories used for training. We propose VACE, Validation-Gated Alternating CoEvolution, which alternates agentic reinforcement learning with trajectory-driven harness refinement. After each RL stage, VACE reuses the collected trajectories to propose a harness revision and evaluates the incumbent and candidate with the updated model held fixed. The candidate guides subsequent training only if it improves validation performance. With Qwen3.5-9B, VACE achieves 45.26% test accuracy on OfficeQA and a mean partial-credit score of 75.19% on AutomationBench, exceeding weight-only RL by 6.43 and 9.09 percentage points and ungated alternation by 4.59 and 6.95 points, respectively. Across 44 harness proposals, 17 reduce validation performance at the updated checkpoint and are rejected before subsequent RL training, highlighting the importance of validation gating.

The rapid, unpredictable advancements in AI system capabilities has seen regulators take adaptive and experimental approaches to policymaking. Established in other domains as instruments balancing regulation with innovation, regulatory sandboxes are seen as solutions for AI regulation. However, analyses mostly focus on the legal and institutional design of AI Regulatory Sandboxes (AIRSes). With the legal framework leaving the socio-technical interpretation to stakeholders, this creates a gap on the sense-making required to fulfill the AIRS purpose. In this paper, we approach this by designing a Boundary Negotiating Artifact as a way to mediate meaning in AIRSes. Through Research-through-Design we iteratively develop a tool, providing an interface for the different stakeholders to collaborate in AI assessment. We then position it as technical backbone in established AIRS frameworks, structuring the collaborative sense-making of the involved stakeholders. We further report the insights gained from our design process leaving the qualitative evaluation for future work.

Accurate perceptual quality assessment is essential for evaluating and optimizing spatial audio, where perceived quality depends on both signal fidelity and inter-channel spatial relationships. However, subjective evaluation is costly, while existing perceptual models are often trained for limited channel configurations and cannot be directly applied to higher-channel-count audio. This raises the question: how can pretrained perceptual knowledge be effectively reused for multichannel spatial audio? Using 5.1-channel audio, we study four levels of multichannel integration: signal, prediction, latent, and feature and propose two learned approaches: latent-level aggregation of spatial-group representations and the feature-level Feature-Band Group Attention (FGAtt), which adaptively fuses spatial groups at the feature level before perceptual processing. Across five 5.1-channel test sets, FGAtt achieves the strongest over- all performance, demonstrating the effectiveness of feature-level adaptation for reusing pretrained perceptual knowledge

Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instability and memory overhead. Even where a critic is trained, it is discarded once training ends, although it has learned to predict outcomes. We revisit this trend and show that a pretrained critic's ability to predict future outcomes can make it a valuable asset for efficient long-horizon reasoning. First, we find that instability in critic-based RL for long chain-of-thought reasoning is largely an optimization artifact: keeping policy updates small and low in variance restores stable convergence. Second, a well-pretrained critic estimates the posterior probability of eventual success from later trajectory states and unfinished prefixes. Its predictions provide outcome-derived, dense, per-prefix learning signals that, during policy optimization, require neither completed rollouts, step-level annotations, nor external reward labels. Building on this insight, we introduce Reward-Free Policy Optimization (RFPO), which repurposes a single calibrated, frozen critic as a rollout-level reward, a value baseline for generalized advantage estimation, and a success forecaster for unfinished prefixes. We further show that binarizing the debiased score stops the policy from exploiting the critic's length bias. Binarized, RFPO matches supervised PPO without a single label in the training loop, while cutting compute and memory overhead. This makes RFPO well suited to long-horizon reasoning tasks, where outcomes arrive late and generation dominates cost: because rollouts can be rewarded before they finish, training no longer has to pay for waiting on every trajectory to complete. Our findings challenge the prevailing critic-free paradigm and establish critic-based, reward-free optimization as a scalable and computationally efficient path for LLM post-training.

Frozen language models (LMs) are increasingly used as fixed feature extractors for downstream reranking, scoring, and preference modeling, raising a practical question: how should a compact module represent interactions among features in a fixed low-dimensional bottleneck? Common linear and low-rank adapters remain linear at the adaptation module itself, whereas explicit second-order alternatives introduce pairwise interactions through direct parameterization or predefined factorizations. We propose SQUARE, a Structured QUAntum REpresentation adapter that amplitude-encodes the bottleneck vector, applies a parameterized quantum circuit, and measures the resulting state. We show that each basis-probability feature is exactly a normalized quadratic form in the bottleneck coordinates, while the additional Pauli-$Z$ readouts are signed linear combinations of these probabilities. The measured map can therefore parameterize interactions over $O(d^2)$ coordinate pairs through a small set of shared circuit parameters, where $d$ is the bottleneck dimension. It provides a structured parameterization within, rather than beyond, the classical normalized-quadratic feature class. In a disjoint same-pipeline evaluation over eight GLUE-derived controlled interaction tasks and five shared seeds, SQUARE achieves an average test accuracy of $0.7565$, compared with $0.7355$ for an affine normalized-quadratic predictor, $0.7271$ for the evaluated parameter-matched Givens mixing model, $0.6817$ for an MLP, and $0.6155$ for a frozen-circuit control. Under reduced supervision, it also shows consistent gains over the strongest evaluated classical comparator, with the same qualitative pattern across multiple frozen LM backbones. All circuit experiments use simulation, while the learned feature map can be evaluated exactly in batched PyTorch without quantum hardware.

World models enable agents to learn and plan in imagination, but predictions beyond their experience can become unreliable and mislead decisions. Existing uncertainty estimates derived from predictions can remain overconfident on unfamiliar state-action pairs. We propose the Lucid World Model (LucidWM), which learns doubt from experience and propagates trust through imagination. By integrating Subjective Logic into categorical latent transitions, LucidWM distinguishes predicted outcomes from their evidential support and assigns each transition a degree of doubt. The complement of this doubt defines transition-level trust, which accumulates multiplicatively along imagined trajectories to reweight returns for policy learning and guide action selection. Uncertainty estimation requires no additional parameters or forward passes. Evaluated on four base world models against seventeen uncertainty readouts, LucidWM detects environmental changes and signals uncertainty during action-corrupted rollouts. In a controlled navigation case study, acting on trust reduces the number of steps required to reach the goal from 362 to 190. Fifteen demonstration videos show how LucidWM doubts its dreams and acts on that doubt. Videos are available at https://lucidwm.github.io.

Mid-training equips pretrained large language models with specialized and reasoning capabilities, but the returns of this stage are bounded since additional serial compute yields little further downstream improvement and can even degrade some capabilities, which places a practical ceiling on how much compute mid-training absorbs. We revisit how this compute should be allocated to a single run or multiple similar optimizations. We find that branches forked from a shared checkpoint under various controlled recipe reaches measurably different regions of parameter space, and establish a form of compatible diversity that extending one run cannot supply. Therefore, we introduce Trajectory Soup, which distributes a mid-training budget over several independent branches, and consolidates strongest checkpoints selected on validation through intra- and inter-trajectory averaging into a single model. A local bias and variance analysis separates the two averaging levels, showing that inter-trajectory averaging removes residual error beyond the reach of averaging within a trajectory, while checkpoint selection carries a bias that bounds how many checkpoints are worth merging. Across model scales, learning-rate schedules, token budgets, and trajectory counts, Trajectory Soup improves aggregate downstream performance over the strongest single-trajectory average under matched budgets and keeps improving as budgets expand, with the advantage preserved after an identical post-training pipeline. These results position trajectory allocation and merging as a practical way to extend the compute-scaling frontier of mid-training beyond serial saturation.

Off-policy and on-policy distillation have traditionally been formulated as separate paradigms, each favoring a different property of distillation trajectories. Teacher-generated (off-policy) traces are typically high-quality but lie far from the student's distribution, whereas student-generated (on-policy) rollouts are more learnable but often contain erroneous reasoning. We view these paradigms as the endpoints of a policy continuum and posit that a more effective rollout policy may lie in between. We introduce \textbf{Interpolated Policy Distillation (IPD)}, which defines the next-token distribution at every decoding step as an explicit linear interpolation between the student and teacher distributions. The interpolation operates at the distribution level, token by token, and its coefficient provides direct control over the balance between trajectory quality and student learnability. Naively sampling from this policy would require sequentially querying the teacher at every token and is thus expensive. To make IPD practical, we accelerate it with a new speculative-decoding rule while exactly preserving the interpolated next-token distribution.At the trajectory level, the resulting rollouts naturally interleave student- and teacher-generated segments. Unlike recent heuristic segment-interleaving methods, however, this interleaving is induced by an exactly realized token-level interpolated policy rather than by hand-designed switching rules. Across text-only and multimodal reasoning benchmarks, IPD consistently outperforms both endpoint policies (SFT and OPD), their conventional two-stage combination (SFT-then-OPD), and recent heuristic segment-interleaving methods, demonstrating that token-level policy interpolation better balances trajectory quality and student learnability.

Human demonstrations capture diverse scenes and rich whole-body skills without requiring robot teleoperation. Prior work on egocentric transfer has emphasized scene generalization in loco-manipulation under decoupled control, leaving direct transfer of coordinated whole-body skills less explored. We present EgoHumanoid-V2, the first egocentric human-to-humanoid skill transfer framework for coordinated whole-body loco-manipulation. At its core, coarse-to-fine action alignment combines kinematic reference correction with dynamics-aware refinement. It improves end-effector pose accuracy while preserving whole-body coordination. We also use robot-arm rendering and training-time image augmentation to reduce the visual embodiment gap and improve viewpoint robustness. On four real-world tasks, vision-language-action (VLA) policies trained on aligned human data show zero-shot skill transfer without target-task robot demonstrations. Task scores are comparable to those of policies trained on teleoperation data at a lower collection cost. These results support human data as direct skill supervision.

Tool-using LLM agents remain vulnerable to indirect prompt injection because trusted instructions and untrusted observations share one context, allowing malicious content to steer consequential input-filtering defenses. Multi-path consensus defenses still leave a high attack success rate because they examine content or aggregated outputs rather than authorizing effects, especially for the within-tool attack, which preserves the intended tool but manipulates its arguments. Data-Flow Control such as CaMeL provides stronger guarantees, but incurs substantial time latency that limits practical deployment. We introduce ToolFence, which compiles a typed authorization blueprint before execution, enforces it through a deterministic monitor, and when the blueprint is incomplete asks a judge to grant new capabilities rather than adjudicate each concrete call. ToolFence provides two key advantages. First, its fine-grained provenance-aware authorization enables the system to distinguish user-authorized values from untrusted observations, effectively addressing the within-tool attack. Second, its deterministic fast path and capability-level runtime grants substantially reduce the frequency of expensive judge calls, improving runtime efficiency. On AgentDojo with Qwen3-max, ToolFence reduces overall ASR to near zero with only a 3.80 percentage-point clean-utility drop and practical runtime overhead.

Large language models can generate plausible code-review comments, but such comments may contain technically incorrect claims that mislead developers. We study technical trustworthiness judgment: determining whether a review comment's core technical claims are correct and applicable to the reviewed code in its repository context. Existing code-review benchmarks primarily evaluate review generation, issue discovery, or general comment quality, but do not directly assess whether an agent can determine the technical trustworthy of an individual review comment. To fill this gap, we introduce CRJudgeBench, a benchmark of 1199 instances constructed from real pull requests and expert-verified perturbations, covering both trustworthy and plausible but untrustworthy comments. We further present Sentinel, a repository-grounded agentic judge that actively gathers code evidence to verify review comments before making judgments. Starting from Qwen3-Coder-30B-A3B-Instruct, Sentinel is trained on the CRJudgeBench training split through iterative action-level learning from a privileged teacher. On the 359-instance CRJudgeBench test set, Sentinel achieves 76.60\% accuracy, outperforming GLM-5.3 by 6.13 percentage points and its base model by 19.78 points. These results show that even state-of-the-art general-purpose LLMs struggle to identify untrustworthy comments, while iterative action-level learning substantially improves the accuracy of repository-grounded trustworthiness judgments. Our dataset is available at https://huggingface.co/datasets/dcloud347/CRJudgeBenchmark

Semantic watermarking improves robustness against watermark removal attacks by embedding detectable signals into sentence-level representations. However, existing watermarking methods typically impose watermark-specific semantic preferences on generated sentences without explicitly accounting for the highly non-uniform and context-dependent semantic preference of LLM generation. When these two preferences are poorly aligned, many natural continuations become incompatible with the watermark, causing semantic narrowing: reduced semantic freedom, increased resampling cost, and potential degradation on tasks with strict semantic requirements. To alleviate this problem, we propose HammingMark, which uses the semantic hash of the preceding sentence as a dynamic center and accepts candidates whose hashes fall within its Hamming neighborhood. Defining watermark validity over a Hamming neighborhood in compact hash space retains a larger fraction of naturally likely semantic continuations. The coarse many-to-one hash mapping further allows diverse semantic realizations to remain watermark-valid. Experiments on C4 and BookSum show that HammingMark achieves strong robustness, high detectability, and near-unwatermarked generation quality, requiring only 2.2 sampled candidates per accepted sentence,a 72.8% reduction compared with the most sampling-efficient existing method. On more complex tasks with strict semantic constraints, HammingMark achieves the highest detection rates with the highest or tied-highest ROUGE-L scores, demonstrating its effectiveness in balancing watermark detectability and generation quality under constrained generation settings.

Answering questions and completing tasks over large document collections often requires connecting evidence spread across multiple documents, such as a project's approval recorded in one, its requirements in another, and its latest status in a third. Recent LLM agents approach this by iteratively searching the full corpus rather than reading only a fixed set of top-ranked documents. However, when the corpus is exposed only as a flat collection of files, a relevant document gives no indication of how it relates to others, so the agent must rediscover these relationships for every query, often missing complementary evidence while simultaneously consuming substantial additional tokens. To address this, we introduce CorpusMap, a navigation layer that organizes the corpus around its recurring entities, which are identifiable from the documents themselves and can link a single document to many others across sources. Specifically, CorpusMap represents each recurring entity as an Entity Page that aggregates information about it and links to every document that refers to it, forming a graph between entities and documents that the agent can traverse to gather otherwise disconnected evidence. Moreover, since CorpusMap is constructed offline by resolving mentions of the same entity across documents, its links are shared across queries rather than rediscovered repeatedly at inference time. Using 7 different models with 3 benchmark datasets, we show that CorpusMap improves both evidence discovery and answer quality over raw-corpus agentic search while using fewer tokens on average, and further outperforms 4 alternative navigation layers, suggesting that entities serve as effective anchors for navigating large document collections.

Datalog underpins reasoning tasks such as program analysis, but its programs are hard to write. Existing synthesizers automate this task but require users to state their intent as input-output examples. Large language models (LLMs) suggest a more natural route, text-to-Datalog synthesis from a natural-language question, yet how well they do so has not been systematically evaluated. We present DatalogBench, a benchmark of 136 text-to-Datalog synthesis tasks curated from existing Datalog-based artifacts. Synthesized programs are graded by execution on held-out inputs against an oracle validated by mutation analysis. Across six LLMs and four prompting configurations, exact match peaks at 68.4%, and relation descriptions or an input-output example have only modest, model-dependent effects. Under direct prompting, most failures occur at compile time, typically because a model invents auxiliary predicates that it never declares or types consistently. Two coding agents reach up to 83.8% and eliminate nearly all such failures, leaving mostly semantic errors concentrated in recursive tasks. DatalogBench thus identifies recursive reasoning and decomposition as open challenges for current LLMs and agents, and offers a reliable, execution-grounded measure of both.

Time series foundation models (TSFMs) are pretrained on heterogeneous collections containing billions of observations, yet their training windows are typically sampled without estimating whether they provide useful learning signal. We introduce a static data-selection framework that scores each window with a reference forecaster and retains an intermediate interval within every source dataset. Specifically, we connect forecasting loss to optimization difficulty by showing that normalized squared loss controls the per-sample gradient norm under a local Jacobian condition. We then define a reference loss score and apply dataset-stratified selection to preserve the diversity of samples. Across various TSFM architectures, our method outperforms random selection by an absolute margin and even improves both relative MASE and CRPS over full-data pretraining by retaining fewer candidate pretraining windows. Further analyses show strong cross-scale and cross-architecture score correlations, indicating that a small reference model can often select data for larger targets, provided that the reference and target share compatible difficulty orderings.

Can reasoning models trick chain of thought (CoT) monitors and perform hidden computation without revealing it in their thinking traces? We show that the answer depends on the underlying task difficulty and the model size. Simple computations can be performed covertly; however, beyond a threshold depending on model size, successfully solving the task necessarily leaks a near-linear amount of information about the covert task input into the CoT. Therefore, sufficiently complex hidden computation always leaves an information-theoretic footprint. However, concerningly, this leakage need not be readable: Under plausible cryptographic assumptions, even a one-layer Transformer can encrypt its reasoning online so that no polynomial-time monitor can extract information about the hidden computation. Overall, our theoretical and empirical results provide a holistic view of both the opportunities and the limitations of CoT monitoring.

Tool-using agents are entering settings where a wrong action carries real cost, and the benchmarks certifying them grade what each simulated tool call reports having done, assuming the tool did what its interface advertises. The audit taxonomies we survey publish no category for that assumption, and a defect beneath a score is present on every rerun. We treat a tool's advertised surfaces as an executable contract, check the implementation against it, and trace each score's provenance through the task files and evaluator code to the verdicts that derive from state a defective tool should have written. Across 34 audited mutating tools in four benchmarks we confirm seven tool defects and one evaluator property at pinned commits. On injected defects the checker raised no false positive in 25 flags, flagged 2 of 5 negative controls, and missed most: in 29 of 33 scored misses a clause covered the defect but no probe revealed it. The checker's own static half, run alone, flags 14 of 17 confirmed sites, so on these findings the dynamic half confirms and traces rather than discovers. Twelve further AgentDojo tools, with six held-out tools and the seven audited first, complete its 25-tool mutating surface, on which at least 5 tools diverge from their advertised surface as our contracts read it, a rate for AgentDojo alone. No gold trajectory reaches either tau2-bench defect; on 1,120 paths built to isolate the telecom defect, a number fixed by construction, the evaluator rewards a refuel of a suspended line and fails the repaired tool. The clearest case is a clinical benchmark whose tool tells the agent each write executed under a documented no-write design its interface does not disclose; its grader takes that message as evidence, so its action success rate records whether a request carried the expected payload, not whether any record changed.

Power markets are a natural testbed for multi-agent reinforcement learning (MARL), where multiple self-interested participants repeatedly submit bids. A market-clearing mechanism then determines dispatch and prices subject to power grid constraints and market settlement rules. However, existing MARL environments typically focus on a single market setting, implement simplified clearing mechanisms, or rely on CPU-based optimization solvers that slow large-scale training and limit the systematic study of bidding strategies and market behavior. We introduce PowerMarketJax, a benchmark suite for MARL across five power markets: day-ahead wholesale, real-time balancing, ancillary services, peer-to-peer double auctions, and local flexibility. Each environment implements its own clearing, pricing, and settlement rules while providing a common framework for learning and evaluation. We find that learned bidding behavior depends strongly on the market design: independent learners can miss better strategies when gains require many agents to change together, when more profitable strategies lie beyond a region of lower profit, or when profits disappear as more agents adopt the same strategy. PowerMarketJax implements both market simulation and policy training in JAX, allowing the entire pipeline to run on the GPU with 1,024 X 1,200 parallelisms across both environments and market participants, achieving up to 33X speedup over CPU-based baselines. Our open-source benchmark is available at: https://github.com/powermarketjax/PowerMarketJax.

Data reconstruction attacks have empirically been successful in recovering training samples from learned models, raising privacy concerns and motivating defenses with guarantees that remain valid against future threats. While differential privacy (DP) provides formal protection, choosing the privacy budget remains a challenge: small budgets severely reduce utility, but it is hard to quantify how large the budget can be without allowing accurate reconstruction. In this work, we study informed attackers who aim to reconstruct a single $d$-dimensional training sample from a $\rho$-zero-concentrated DP model, knowing all other training data. Our main contribution is to establish a sharp transition at $\rho \asymp d$ for data reconstruction: on the one hand, we derive entropy-based lower bounds for any private mechanism and any attack, characterizing a set of target priors for which reconstruction is information-theoretically impossible for $\rho \ll d$; on the other hand, we analyze a simple attack on private linear regression with output perturbation, showing that reconstruction is practically feasible for $\rho \gg d$. Remarkably, the transition moves to $\rho \asymp s$ for data lying in an $s$-dimensional subspace, demonstrating that the privacy budget guaranteeing adequate protection must be assessed in terms of the effective dimension of the data. We validate our findings via experiments on synthetic data and natural images (CIFAR-10, ImageNet).

Test-time compute scaling has emerged as a cornerstone of advanced machine reasoning, yet performing iterative deliberation directly within continuous latent representation spaces reveals a catastrophic pathology: the Deliberation Drift Cliff. While unconstrained recurrent latent models achieve initial reasoning gains at short horizons (K <= 4), their reasoning collapses when extrapolated to deeper thinking steps (K >= 16), dropping by 22% to 62% across standard logical benchmarks. We resolve the trilemma among expressivity, Lyapunov stability, and computational efficiency in test-time latent reasoning through a 22-round empirical and theoretical investigation. We demonstrate that strictly conservative scalar potential gradient flows suppress long-range drift (cliff 3.40%) but bottleneck peak reasoning accuracy at 32.73%, whereas unconstrained rotational flows achieve high symbolic expressivity (82.33%) but suffer a severe 36.87% drift cliff. To resolve this geometric duality, we establish Port-Hamiltonian Latent Deliberation (PH-LD) and propose the Direct-Gradient Pure-Tensor Helmholtz-Hodge Decomposition (DG-HHD). DG-HHD parameterizes the attracting flow as a tangent projection tensor network while orthogonally decoupling non-zero circulation (Hodge machine error 1.65e-17, contraction error 5.55e-17), eliminating runtime autograd dependencies to achieve 1.84x vector field and 2.09x RK45 rollout speedups. In a 15-arm symmetrical Pareto benchmark, DG-HHD achieves 58.67% peak accuracy (+25.94% absolute gain over conservative HHD) and retains 35.27% at K=32. Transferred to small language model (SLM) multi-hop causal reasoning, DG-HHD delivers monotonic compute scaling (49.33% to 51.56%) and suppresses out-of-distribution drift (cliff -0.66%). All 30 Level 0 deterministic invariants are certified.

Model adaptation is typically governed by a fixed recipe, even though different update programs can produce substantially different behavioral outcomes. We introduce adaptation compilation, which reframes where, how, and to what extent a model should adapt as a joint prediction and decision problem. Rather than searching over candidate programs anew for each learning episode, a compiler learns from prior adaptations to predict a vector-valued counterfactual response surface over candidate programs---their expected effects on acquisition, transfer, boundedness, and preservation---and selects a program before adaptation begins. Because this predicted geometry captures multiple behavioral consequences rather than a single winner or scalar score, it can be reused under different downstream priorities without retraining. Across five learning types, preferred programs vary meaningfully across episodes, and this variation is predictable from pre-adaptation information. On Llama-3.1-8B, compiler-selected programs approach exhaustive search while outperforming global and objective-specific defaults. Replication on Gemma-2-9B preserves program heterogeneity and selection headroom, but shows that exploiting this headroom requires accounting for uncertainty when departing from strong defaults. Together, these results show that adaptation search can be amortized across related learning problems, turning prior adaptation experience into a basis for deciding how future learning should occur.

Large language models (LLMs) are increasingly used for software engineering tasks that require understanding existing source code, including behavior prediction, function explanation, debugging, and code review. However, aggregate benchmark accuracy can conceal how model reliability changes as source code becomes structurally more complex. This paper presents a complexity-aware framework for evaluating LLM code comprehension using cyclomatic complexity, nesting depth, branching factor, and Halstead volume. We evaluate DeepSeek-Coder-V2 and Llama through two complementary tasks: automatic input-output prediction over 300 Python functions and manually assessed semantic comprehension over a balanced subset of 60 functions. The functions are grouped into Low-, Medium-, and High-complexity bands. DeepSeek-Coder-V2 achieves an overall automatic accuracy of 78.33%, compared with 70.33% for Llama. However, accuracy decreases substantially from Low to High complexity, from 93.52% to 52.78% for DeepSeek-Coder-V2 and from 87.04% to 47.22% for Llama. Incorrect predictions are consistently associated with higher values of all four complexity metrics, and correlation and logistic-regression analyses confirm broadly comparable negative associations between structural complexity and correctness. Manual semantic comprehension shows the same degradation pattern, with accuracy decreasing from 100.00% to 75.00% for DeepSeek-Coder-V2 and from 90.00% to 60.00% for Llama. These findings demonstrate that complexity-aware evaluation provides a more diagnostic assessment of LLM code-comprehension reliability than aggregate accuracy alone.

Optimal Transport (OT) provides a principled framework for learning transformations between probability distributions from unpaired samples. In many applications, however, a single transformation must map several source distributions to a common target distribution. For example, image restoration might require handling different types of degradation without knowing the degradation of each input at inference time. Simple approaches of pooling the source distributions only encourage alignment with the target at the aggregate level and may leave individual sources misaligned. In our paper, we consider the simultaneous OT problem which formalizes the task of learning a shared transport map that minimizes the average transport cost while aligning each source distribution with a prescribed target. We propose a neural method for solving the simultaneous OT problem by learning a shared transport map that minimizes the average transport cost while aligning each source distribution with a prescribed target. We derive a max-min formulation for learning this map. We illustrate its application to image restoration, where a single model handles multiple degradation types using a common collection of clean target images.

Long-term memory enables language models to use past interactions in future conversations. However, evidence needed to answer a question may be scattered across distant turns, while the question itself omits clues needed to locate it. Retrieved facts can reveal these clues, motivating retrieval decisions conditioned on evidence already found. We introduce MERA (Missing-Evidence Retrieval Augmentation), which separates globally searchable memory from a question-specific evidence state. Verified evidence guides subsequent retrieval without restricting access to the global memory. We train a lightweight planner through reinforcement learning, rewarding queries that recover previously missing evidence. MERA achieves strong answer accuracy across Qwen3-30B and GPT-4o-mini backbones. With Qwen3-30B for evidence processing and answer generation, the trained 0.6B planner achieves 77.40% accuracy on LoCoMo and 71.29% on LongMemEval-S, exceeding a 30B planner without retrieval-grounded training by 4.10% and 3.96%, respectively. On LoCoMo, later retrieval rounds increase cumulative evidence recall from 55.5% to 80.5%.

How strong can an AlphaZero-style chess system become under limited training compute when its entire learning loop is engineered for efficiency? We train from random initialization through searched self-play on a single eight-GPU node for 2.5 days. The resulting 6.32-million-parameter model reaches 3,251 benchmark Elo [3,206, 3,297] at 100,000 searches per move (estimated at under five seconds of thinking time) against a fixed-node Stockfish 13 ladder. The run ingests 3.25 million completed games, involves an estimated 100 billion search simulations, and makes 836.6 million training presentations. We investigate search allocation, replay and restart-state selection, policy representation, progressive model sizing, quantized inference, and throughput engineering. Alongside the retained design, we document plausible alternatives that failed to improve the complete learning loop or did not justify their cost. The reported strength is a result of the integrated system, not an isolated Elo gain attributable to any single choice.

Agentic retrieval-augmented generation (RAG) interleaves reasoning with repeated retrieval, giving the retriever influence over both the evidence an agent observes and its subsequent search decisions. We study retriever backdoors that exploit this feedback loop and repurpose weak backdoor purification to conceal their presence. An attacker supplies a compromised retriever checkpoint while leaving the search agent and deployment corpus unchanged. Without corpus write access, the attacker can still suppress useful evidence, persistently retrieve a selected existing document, or steer the agent toward prolonged search, inflating retrieval, context, and latency cost. To conceal these behaviors from detection, we propose leveraging a controlled inject-and-remove cycle: deliberately inject a weaker backdoor and then unlearn it. This process weakens detector-visible signatures and fools the backdoor detectors with an illusion of purification while preserving the malicious retrieval behavior. These findings expose a systematic vulnerability in RAG systems in which a weak defense becomes an attacker's concealment tool for a backdoored retriever, even when the underlying corpus remains trustworthy.

Retrieval-augmented language models are expected to answer from the retrieved evidence, but in practice they often keep answering when that evidence is missing. We trace this behavior to the training signal: answer-focused fine-tuning assigns no target to unsupported contexts, so it cannot distinguish a reader that abstains from one that guesses, and unsupported answering stays near 100% even as supported accuracy improves. We introduce Regime Boundary Alignment (RBA), which trains a single reader on matched variants of the same question and gold answer. The reader is trained to produce the gold answer when the context supports it, including when conflicting evidence is also present, and to abstain when the correct support is removed; inference is ordinary decoding, with no verifier, threshold, or regime label. On three multi-hop QA datasets across three seeds, RBA reduces the unsupported-answer rate by more than sixty percentage points relative to conflict-focused training while matching its supported accuracy. On a held-out TriviaQA retrieval-miss slice, the same reader reduces unsupported answering from 100% to below 1% while also improving supported accuracy. These results indicate that evidence-gated answering must be learned on both sides of the support boundary.

Serving an answer from a large language model requires deciding when to abstain, yet a verifier's ranking accuracy alone does not determine the error rate among served answers. We introduce PriceCheck, which builds a compact family of decision rules from label-free checks such as re-solving a problem. Each check has a price: its agreement rates on correct and incorrect answers and its cost per run. Prices fitted on a small, class-enriched labelled set compose into predictions of a schedule's coverage and cost, guiding which checks to run and when to stop. A calibration test then selects a schedule at a stated selective-risk target. In mathematics, the selected schedules serve 76.1% of answers on average and keep held-out selective risk below 1.5% on all 15 splits. Under the shared testing protocol, PriceCheck serves more answers at that target than reward models, a prompted judge, the generator's confidence and a trained correctness classifier. At matched coverage, it keeps the fewest wrong answers among these scorers. Across 118 diagnostic schedules, price-based coverage predictions have a rank correlation of 0.97 with observed coverage. These results show that choosing how checks are combined and stopped matters alongside how well a verifier ranks answers. Code is available at https://github.com/js-lee-AI/PriceCheck.

We propose RegLLM, a diagnostic harness for bounded autonomy in regulated agentic workflows. It instruments six trustworthiness signals: citation validity, source grounding, schema compliance, escalation correctness, constitutional alignment, and unsafe-action rate. Signals are distinguished by their source of supervision: programmatic verifiers, task-level escalation labels, or AI-judge scores. A deterministic runtime supervisor blocks ungrounded answers and forces escalation, logging interventions. The same domain constitution informs evaluation, training rewards, and serving guardrails. Task-level should-escalate labels make the act-versus-defer decision a measurable training signal. We demonstrate the harness at smoke scale. An offline reference run (n=12) lifts escalation recall from 0 to 0.67 and reduces unsafe-action rate from 0.33 to 0.08 when governance is enabled. Two single-GPU Qwen2.5-3B LoRA/DPO pilots (n=8, same seed and evaluation split) expose substantial variation: nominally identical RL-base configurations yield task success of 0.25 versus 0.12 and escalation recall of 1.0 versus 0.5. An answer-quality adapter changes recall from 1.0 to 0.5 in Run A, but from 0.5 to 1.0 in Run B. An escalation-aware variant produces no measurable change in Run B. These small pilots do not establish reliable adapter effects or production readiness. Their contribution is diagnostic: configuration variance can overwhelm apparent tuning effects on bounded-autonomy metrics, motivating larger evaluation sets and repeated runs.

Video-based policy learning is particularly promising, as it illustrates target behaviors without requiring action annotations or embodiment-matched demonstrations. A central challenge is deciding what information should be transferred from the video to the robot. Existing approaches commonly convert visual observations into scalar similarity or value signals, or ask foundation models to directly generate reward code. These approaches can make the temporal structure of a task difficult to inspect, ground, and reuse. We present Video2STL, a framework that converts observation-only videos into parametric Signal Temporal Logic (STL) specifications and uses the resulting formal representation for robot learning. A vision-language model extracts an embodiment-independent semantic event trace and constructs a bank of symbolic temporal specifications. The model determines the task structure, while numerical predicate thresholds and temporal bounds are grounded from successful robot trajectories. For policy learning, we separate short- and long-timescale temporal information: short-horizon specifications provide dense rewards through rolling-window quantitative robustness, while a causal monitor over a retained long-horizon specification provides one-time progress rewards for valid temporal prefixes. The same representation supports cross-embodiment transfer from human or animal videos to robot control. Across four manipulation tasks, Video2STL achieves $85.8\%$ average success-once and $67.0\%$ success-at-end, compared with $81.5\%/59.5\%$ for native dense PPO and $65.0\%/42.3\%$ for Text2Reward; in quadruped locomotion, Qwen-3.8 and GPT-5.6-based Video2STL policies achieve $100\%$ success across velocities from $0.3$ to $2.1\,\mathrm{m/s}$ while remaining competitive in high-speed energy efficiency. Project webpage: \href{https://video2stl.github.io/}{video2stl}.

Growing large language model applications demand efficient inference. At high concurrency, block-diffusion speculative decoding suffers from verification padding, rejected candidates, and incompatibility between variable prefixes and fixed-shape graphs. Uniform truncation sacrifices acceptable tokens. We present DScale, preserving drafter architecture, weights, and full draft length. A separate 112K-parameter predictor requires neither confidence calibration nor hardware speed-curve preparation. Path-aware tiles reduce padding. Dynamic verify-length (DVL) allocation packs scored prefixes into half the native verification capacity. Fixed-address workspaces propagate changing boundaries through verification and acceptance while reusing captured graphs. On A100-40GB with tensor parallelism 1, Qwen3-8B and Qwen3-4B cover four datasets and concurrency 8-32, reusing each target's frozen predictor. Geometric-mean throughput gains across these configurations are respectively 43.9% and 48.8% over DFlash, 22.2% and 37.7% over DSpark, and 24.4% and 32.0% over Domino, with lower request latency. Cumulative ablations show that adding the three mechanisms successively increases geometric-mean throughput, while budget adjustment improves accepted-token retention. GPU profiling shows that complete decode-step time on GSM8K decreases by 30.8-52.5% relative to DFlash

Large language models (LLMs) are increasingly used as high-level planners in robot navigation, but their outputs may become unreliable when instructions are ambiguous, unsupported by the environment, or semantically inconsistent. This paper presents a Risk-Aware Semantic Grounding framework for trustworthy LLM-based robot planning. Unlike existing LLM-based planners that primarily optimize plan generation, we formulate semantic grounding reliability as a multi-dimensional risk estimation problem. The proposed architecture explicitly models grounding uncertainty through ambiguity, hallucination and semantic-conflict risks before planning occurs, enabling the system to decide whether to execute the instruction, request clarification, or reject it. To evaluate the approach, we introduce TRUST-NAV, a benchmark containing both standard navigation tasks and risk-inducing instruction scenarios. Experimental results show that while conventional LLM planners achieve strong performance on valid navigation tasks, the proposed framework substantially improves ambiguity detection and semantic conflict rejection. These findings suggest that trustworthy robot planning should be evaluated not only by task completion, but also by the ability to recognize when execution should not occur.

Driven by the rapid advancement of large language models (LLMs), LLM-based multi-agent systems (MAS) have emerged as a powerful paradigm for collaborative reasoning over complex tasks. A key design element of MAS is the communication topology, which governs information flow among agents and often encodes proprietary knowledge about the system architecture. However, recent work has shown that such topologies can be inferred even in black-box settings by exploiting semantic dependencies in observable reasoning traces, posing significant risks of intellectual property leakage and exposure of system vulnerabilities. To address this threat, we propose MIRAGE, a topology-concealment framework that preserves the genuine communication topology for task execution while shaping adversary-facing semantic evidence toward a carefully constructed phantom topology. Specifically, MIRAGE operates in three stages: (1) phantom topology synthesis, (2) semantic edge realization, and (3) protected MAS execution. It constructs a phantom topology structurally distinct from the genuine one, materializes phantom edges as plausible semantic dependencies, and suppresses source-specific cues that could reveal genuine edges absent from the phantom topology. Extensive experiments across three topology optimization frameworks and four benchmark datasets demonstrate that MIRAGE substantially reduces the effectiveness of topology inference attacks while largely preserving the task utility of the protected MAS.

Continual audio deepfake detection requires learning newly emerging deepfake methods while retaining discrimination of previously encountered speech. Existing dataset-incremental evaluation changes both real-speech domains and deepfake mechanisms, making their effects difficult to distinguish. We construct five task organizations over identical training, development, and evaluation pools to study these factors under a controlled sample budget. Our proposed Real-Anchored Mechanism-Incremental (RAMI) protocol reflects the practical setting in which available real speech provides a recurring mixed-domain reference while new deepfake mechanisms arrive incrementally. We further propose RF-Prompt, an asymmetric continual prompt-learning method that preserves reusable real-speech knowledge through a shared real prompt and expands mechanism-specific knowledge through inherited fake experts with orthogonal residuals. Input-adaptive soft fusion combines the accumulated experts into a fixed number of injected tokens without requiring task identity at inference. On RAMI, RF-Prompt achieves 10.110% average EER and 10.370% pooled EER, outperforming all evaluated continual-learning baselines. Across the five controlled protocols, RAMI yields the lowest common-average and pooled EER. Component ablations, limited-data experiments, and cross-backbone evaluations further validate the proposed design.

Longitudinal electronic health record (EHR) modeling requires integrating new visits with an expanding patient history. Yet the continual accumulation of clinical information imposes increasing computational and memory costs on large language models (LLMs) when they process and retain complete patient histories. A practical alternative is visit-wise recurrent compression, which incorporates each incoming visit into a compact, continually updated patient memory. However, under a fixed memory budget, successive updates must integrate new information without progressively losing critical historical evidence needed to subsequent tasks. To address this challenge, we introduce Recurrent Longitudinal Memory (ReLMem), a framework that learns to maintain fixed-capacity patient memory for efficient downstream prediction with a frozen LLM. ReLMem equips this LLM with lightweight compression adapters to recurrently update the memory from its previous state and each incoming visit, without rereading earlier records. Specifically, we develop a multi-granularity optimization strategy to preserve task-relevant information throughout recurrent updates and support downstream prediction from the final memory. The intermediate supervision aligns attention outputs from compressed memory and the full history under identical queries, while prediction supervision minimizes cross-entropy with ground truth answers conditioned on the final memory. On EHR-based medication prediction, ReLMem approaches the F1 scores of full-history baseline while reducing average retained historical storage by 97.1%. Under the same memory budget, it improves macro- and micro-F1 over the strongest compressed-memory baseline by 4.66 and 4.75 percentage points, respectively. These results highlight the value of learning recurrent patient memory for efficient longitudinal EHR modeling.

Test-time adaptation for vision-language navigation (TTA-VLN) enables pretrained policies to adapt online to unseen environments using only test-time observations and interaction history. However, distribution shifts can distort local action preferences and lead to off-course decisions. Existing methods rely on predictive uncertainty, trajectory-level feedback, or accumulated adaptation experience to correct such deviations. These signals, however, do not directly reveal whether an executed action supports instruction-guided progress toward the goal. Moreover, a plausible corrective signal does not guarantee a reliable policy update. The key challenge is thus twofold: identifying interactions that support goal-directed improvement and determining whether the resulting updates are worth retaining. We observe that each executed action induces an immediate observation transition, providing evidence of its local consequences. Based on this insight, we propose Credit-Guided Policy Improvement (CGPI), which recovers signed, reference-relative decision credit from action-induced observation transitions without external outcome feedback. With the pretrained navigation policy frozen, CGPI uses this credit to propose lightweight adaptation updates and verifies them against prior credit-supported interactions. Updates are retained only when supported and rolled back otherwise. CGPI achieves consistent gains across the evaluated VLN benchmarks and navigation backbones, while qualitative robot trials further illustrate the feasibility of zero-shot sim-to-real transfer.

A common safeguard for a data pipeline is redundant computation: derive each published number by two routes built on different technology and refuse to exit when they disagree. We report one such gate failing, in a cross-catalogue integrity study of two open registers of Earth-orbiting objects. A gate comparing a set-based Python path with SPARQL queries over the emitted RDF graph printed ALL CROSS-CHECKS AGREE on seven counts. Three were wrong, one overstated more than fourfold (932 against 220). Both paths imported the same constants, which encoded a misreading of the source's status vocabulary, so the error was common-mode and the gate could not see it. We give the mechanism, an object-level ledger reconciling every figure, and three checks that go back to the source's documentation, measured on the defective code and on its correction. We then checked that correction against each object's phase history, held in a source file the pipeline never read. The correction was also wrong: 42 of its 261 disagreements are artefacts, and none of our three checks flagged them. Finally, in a controlled replication with three pinned models and tools disabled, 72 of 75 paths generated on request as independent checks computed the defective count, 29 of 30 even when the prompt carried the source's own definitions of the codes. The evidence is one pipeline and one defect family. Within it, redundancy verified implementation, and the errors that reached publication were errors of meaning.

Speculative decoding accelerates autoregressive generation by using a smaller drafter to propose tokens for batched verification by a larger target. However, conventional speculative decoding couples drafting to the target's evolving verified prefix, serializing drafting and verification. We ask whether this dependency is necessary for source-conditioned generation. Our key observation is that, for audio language models, the input audio and user request can provide useful speculative candidates without following the target's evolving text prefix. We propose AS$^2$D (Audio Speculative Speculative Decoding), which enables target-decoupled drafting: an audio-conditioned drafter follows its own generation history while the target independently verifies and corrects ready candidates. Without usable candidates, the target advances alone. Thus, target feedback determines which candidates are committed but no longer determines when the drafter can make progress, enabling drafting and verification to proceed concurrently while retaining target-side verification and correction. We implement AS$^2$D in MNN for Android and evaluate two target models across four phones, seven datasets, and three tasks covering 12.2 hours of audio. Across four phones, AS$^2$D improves pooled ASR throughput by 42-76% over target-only decoding, while only 5.7% of evaluation windows are slower than target-only, compared with 58.1-63.0% for speculative baselines. For ASR, AS$^2$D reaches 97.33-98.20% of a hindsight per-window oracle's pooled throughput over the evaluated drafter/budget catalog. Native on-demand execution with a 7B target achieves up to 78% higher throughput than target-only. These results show that source-conditioned audio generation can relax the conventional dependence of speculative drafting on the target's evolving output prefix, exposing substantial parallelism for efficient inference.

Fine-tuning a language model on a narrow set of harmful demonstrations, such as bad medical advice, can make it broadly misaligned on unrelated questions, a phenomenon known as emergent misalignment (EM). The usual defense is to find the offending rows and delete them, but a row locator failed our held-out test and deleting rows helps less than expected. We ask a different question: given a fixed set of poisoned rows, is it better to correct them than to remove them? We fine-tune Qwen2.5-14B-Instruct on a mixture of bad medical advice and benign chat data, select a quarter of the poison rows in advance, and either delete them or replace each with a corrected answer to the same prompt, keeping everything else the same. Replacing the rows cuts the EM rate by about a third and improves answers on held-out medical questions, while deleting the same rows has little measurable effect. The advantage is larger when half the poison rows are corrected, and it holds on a second base model and a second misaligned model organism. The content of the replacement appears to matter: paraphrasing the rows while keeping their bad advice shows no clear benefit, and the correct answers distributed with the dataset appear to do about as well as our rewriter's. Realigning an already-poisoned model with further fine-tuning is known to work, but which data does the work has not been compared directly. We find that a short round of training on corrections beats the same amount of training on generic chat data, that corrections on other medical prompts do roughly as well as corrections of the poisoned prompts themselves, and that instructing the correction writer to model a careful, harm-avoiding assistant adds no measurable benefit over plain corrections. In the settings we tested, correcting harmful training data reduces EM more than deleting it.

No single way of parallelizing attention serves large language models well under all loads. Low concurrency favors tensor parallelism, many independent requests favor data-parallel attention, and long prompts favor context parallelism. Reasoning, agentic, and RL-rollout workloads make a fixed choice untenable: a batch that begins as many short requests ends as a few very long ones, so the best layout changes while the same requests run. Serving engines nevertheless fix one layout at launch, because changing it has meant draining requests and restarting workers. We present SPLASH, a serving system that switches the parallel layout of attention while requests are running. It builds on one observation: modern attention, with few or no KV heads, decouples where a request's KV cache lives from how attention weights are sharded. This has two consequences. First, layouts differ only in who owns the weights and the cache, and most of that state already sits where the next layout needs it; SPLASH reuses it, moves the rest in the background of ongoing inference, and hands off at a batch boundary, making a switch nearly free: its median overhead is under 0.51% of the step it runs in. Second, the decoupling exposes a layout that existing engines lack: Decoupled Ownership Parallelism (DOP) shards attention weights as tensor parallelism does while keeping each request's cache on a single owner as data-parallel attention does. DOP replicates neither, offers 27-60% more KV capacity than data-parallel attention, and gives the scheduler a choice when KV memory limits admission. A transition-aware scheduler follows the best of the four layouts as load changes. On B200 GPUs serving GLM-5.3, SPLASH improves end-to-end serving throughput by 1.3-1.73x over fixed-layout deployments, and the same layout regimes appear with DeepSeek-V3.2 on H200 and GLM-5.3-Flash on DCU.

Time series imputation has progressed from statistical and deep learning approaches to diffusion-based models, which have shown strong recent performance. Existing diffusion-based methods typically condition the reverse process using local contextual information from the current or neighbouring windows. Meanwhile, global dataset-level structure often remains implicit, limiting performance when local observations are sparse, noisy, or unrepresentative. To address this issue, we propose ProCTI, a diffusion-imputation framework that augments local conditioning with retrieved global dataset-level priors through learned prototypes. A hybrid conditioning mechanism integrates this global context with local signals during reverse diffusion, enabling more accurate reconstruction under varying missingness scenarios. Experiments across multiple benchmark datasets show that ProCTI outperforms strong baselines overall under random missingness, while remaining competitive under attribute-wise missingness. Furthermore, we use a latent-regime data model to characterise the precise conditions under which prototype-derived global conditioning provably improves imputation. We support this with a general theoretical analysis of local-global conditioning.

The common paradigm of reinforcement learning with verifiable rewards (RLVR) is to let agents make multiple attempts at a task, and optimize towards the successful ones. This becomes problematic in the realms of self-improvement, where tasks are so difficult that the agent has a low or even no chance of success, and where there are no teacher models or example solutions to distill from. In this paper, we introduce RLTL;DR. After each failed attempt, we show the policy the verifier outputs and let it write its own feedback, in the form of a single TL;DR insight. The next rollout is conditioned on all previous insights, and we sequentially sample rollouts until a solution is found. Moreover, we enable backpropagation on the in-context insights to internalize a direct task to insight mapping. On challenging tool-calling and coding datasets (filtered to Pass@128=0), standard GRPO training of a Qwen 3.5 9B Thinking policy stays flat at a Pass@1 of 0% to 1%. RLTL;DR breaks through this learning barrier, achieving a Pass@1 of 14-31% with insights in context during training and, crucially, 12-13% when no insight is in context at eval time. We identify that the key is the task to insight internalization. To study this further, we reduce our approach to SFTL;DR, training only on (task, insight) tuples, without showing or backpropagating on any rollouts. Training on only 4k of these tuples recovers almost the full performance of RLTL;DR and classical SFT on full rollouts. This demonstrates a promising compacted training paradigm of the form "on this sort of task, keep this sort of thing in mind", which we hope to inspire future research on.

Jev is a commercial System One model from TypeSafe AI that does not generate text: given a state and typed questions, it returns a choice from fixed options, a position on a rubric, or the probability that a statement is true, with probabilities the vendor describes as calibrated. Such models target small decisions in information access pipelines, such as routing queries, checking grounding, moderating content, or rating against a rubric. We evaluate Jev (jev-1.13.0) zero-shot on 37 datasets spanning classification, routing, natural language inference, reading comprehension, commonsense reasoning, moderation, legal clause analysis and rubric scoring, with one frozen template per dataset and full evaluation splits: 346,009 requests for under USD 10. For reference, we score Qwen3.8-27B and Gemma-4-E4B on identical requests via their exact next-token probabilities over the options. Jev reaches 95-99% accuracy on IMDB, SST-2, HellaSwag and ARC and 86.7% on Belebele across 122 languages. It beats Qwen on 27 of 37 datasets, with none of Qwen's nine leads outside the bootstrap intervals, and Gemma on all 37. All three models degrade on low-resource languages, fine-grained or noisy labels, and rubric-based quality judgments. Jev's choice probabilities are well calibrated and support selective prediction. Binary probabilities rank well but are poorly placed relative to a fixed 0.5 threshold; thresholds tuned on training data raise micro-F1 on UNFAIR-ToS from 0.50 to 0.75. Jev answers MMLU's calculation-heavy questions more accurately than other MMLU questions (94% vs. 91%), whereas both open models, and all three on C-Eval, find them harder. Rotating the options leaves Jev's accuracy unchanged and withholding the question drops it to near chance, ruling out shallow memorization but not memorized question-answer pairs. We release the code, harness and all raw responses.

Search and Rescue (SAR) operations increasingly deploy heterogeneous teams of aerial and ground robots. However, conventional coverage methods typically do not translate perceived terrain into platform-specific reachability, while continuous image exchange imposes a high communication cost. We propose an edge-centric, semantic-aware coverage planning framework that integrates aerial terrain perception, robot-specific traversability reasoning, and payload-efficient semantic state sharing. Aerial observations are converted into compact semantic grid maps, enabling reachability-constrained area decomposition and capability-aware coverage paths that assign only regions admitted by each robot's capability profile. The resulting perception-sharing-planning loop feeds semantic corrections into traversability reasoning and replanning, forming an application-level mechanism motivated by AI-enabled goal-oriented communication envisioned for AI-native 6G networks. For the high-update case, transmitting semantic corrections reduces the application payload by a factor of approximately $82$ relative to periodic full-map sharing. Across matched benchmark scenarios, the proposed method achieved $91.5\%$ coverage with no capability-infeasible allocations, compared with $78.8\%$ coverage and a $21.5\%$ capability-infeasible allocation rate for LS-MCPP. Semantic corrections update the shared planning state without requiring repeated transmission of the complete map.

Retrieval-Augmented Generation (RAG) is increasingly used to enhance Large Language Model (LLM)-based software vulnerability detection by grounding predictions in retrieved vulnerability knowledge, such as vulnerability reports. However, existing RAG-based software vulnerability detection (RAG4SVD) systems are often evaluated using proprietary models, which challenges open science and reproducibility. Further, studies use different datasets, custom knowledge bases, different backbone models, and diverse metrics, which hinders meaningful cross-system comparison. In this work, we study six open-source RAG4SVD systems and address these reproducibility and comparability challenges through (i) reproduction of their experimental settings under an open-weight setting, and (ii) a unified benchmark using a common dataset, metric suite, and pool of open-weight models. Further, RAG4SVD systems typically consist of multiple components, yet are often evaluated only as a whole system, i.e., end-to-end. Therefore, we perform (iii) a component-level analysis that decomposes representative RAG4SVD pipelines into input abstraction, knowledge retrieval, and detection. Our results demonstrate that reproducibility varies substantially across systems. Under the presented unified benchmark, published RAG4SVD performance does not transfer under a controlled open-weight evaluation and depends strongly on the used model. The component analysis shows that effective RAG4SVD depends on the alignment between pipeline stages. For example, oracle knowledge raises retrieval to near-optimal, yet performance remains low (0.51 pairwise accuracy), demonstrating that retrieval effectiveness alone is insufficient for reliable detection. These findings motivate evaluating RAG4SVD not only end-to-end, but at the level of pipeline components, and provide a basis for more standardized, RAG-aware evaluation practices.

One of the current premises of mechanistic interpretability research is that detailed accounts of the geometry of neural network representations can tell us how models perform computations, and how to effectively intervene on them. While low dimensional manifolds have been observed for multiple concepts in the literature (e.g. numbers encoded on helices, days of the week on a circle, ...), with structure believed to reflect properties of data and tasks, the extent to which models rely on them for computation, and how they manipulate them, remains unclear. We characterize precisely the geometry of computation in a number-comparison task, as an abstraction of comparison for decision making, and how models utilize geometry in an elegant fashion to implement it. Specifically, we study the causal geometry of number comparison in Qwen2.5-7B-Instruct, a capable and widely studied open-weight model, and find Qwen largely uses linear representations of numbers despite the presence of curved geometry. To compare two numbers, the model first encodes each number along a vector and adds the two representations using attention and the residual connection, bringing them into a shared space in the residual stream. Then, the model uses MLP neurons to compare the pair of numbers on local regions in this shared space, which correspond to smaller intervals of input numbers, and combines these to obtain the position of the maximum. In fact, this reliance on linear representations for comparison also persists when the model compares three numbers. Our findings demonstrate that the manifold hypothesis can co-exist with linear representations: while concepts that are ordered may have manifold structure in representations, the model may use an underlying linear structure of the concept in certain computations.

Diffusion models have recently shown strong potential for probabilistic multivariate time-series forecasting by modeling complex conditional distributions. Recent decoupled diffusion frameworks further separate forecasting into deterministic prediction and stochastic residual generation, making it natural to derive dependency graphs from deterministic representations and use them to guide residual diffusion. However, we show that this direct structural transfer is unreliable. Although deterministic-derived graphs encode useful global dependency priors, they exhibit substantial edge-level misalignment with residual dependency structures, introducing inaccurate or redundant conditions during residual generation. This reveals a previously overlooked deterministic-to-residual structural alignment problem in decoupled diffusion forecasting. To address this problem, we propose GARDiff, a Graph-Aligned Residual Diffusion framework for probabilistic multivariate time-series forecasting. Instead of treating deterministic-derived graphs as fixed diffusion conditions, GARDiff progressively adapts them to residual generation. Specifically, GARDiff estimates residual uncertainty to distinguish high- and low-uncertainty regions, enabling uncertainty-aware structural refinement, and further performs timestep-aware edge sparsification during reverse diffusion to evolve graph conditions from broad dependency aggregation to localized residual refinement. Extensive experiments on six real-world benchmarks demonstrate that GARDiff consistently improves probabilistic forecasting performance and uncertainty calibration over strong baselines.

Class Incremental Learning (Class-IL) requires models to learn new classes over time while preserving previously acquired knowledge without access to past data or task identity. This setting intensifies the stability-plasticity dilemma and makes catastrophic forgetting a central challenge. Existing approaches include regularization, knowledge distillation, replay, and architectural expansion. However, many expansion methods rely on explicit task identifiers or predefined growth strategies, limiting their applicability when task boundaries are unavailable at inference time. This work proposes a dynamic width expansion method that increases the number of neurons within existing layers according to a normalized loss criterion, without requiring task-specific information. An attention mechanism with persistent key-value memory is also incorporated to stabilize feature representations and reduce interference between previously learned and newly introduced classes. The approach is evaluated on Split MNIST and Split CIFAR-100 under the standard Class-IL protocol. Experiments compare fixed-capacity and dynamically expanding architectures, both with and without attention, combined with established continual learning methods including EWC, LwF, and A-GEM. Results show that progressive width expansion consistently improves performance over fixed architectures, particularly when combined with functional methods and A-GEM. The combination of width expansion and attention provides the most consistent gains. Overall, dynamic width expansion based on representational demand provides an effective and flexible strategy for Class-IL, although uncontrolled growth may increase overfitting and computational cost.

Rubrics support the structured evaluation of language models. We propose a rubric for assessing expressed clinical reasoning in model responses, drawing on three bodies of work: medical education assessment frameworks (ART, SCT, Key Feature Problems and OSCE); clinical LLM benchmarks (MedR-Bench, HealthBench, TIMER-Bench, DR.BENCH, PrIME-LLM and PatientSafeBench); and general LLM reasoning evaluation research, including the Factuality-Validity-Coherence-Utility taxonomy, FaithCoT-Bench and C2-Faith. We use groundedness as a clinically oriented adaptation of the taxonomy's factuality category. The rubric brings these concepts together in a multidimensional framework for scoring free-text responses to gold-standard clinical vignettes. It includes provisional behavioural anchors, applicability rules and a separate flag for case-specific safety-critical errors. General-domain frameworks inform its design but are not treated as validated clinical instruments. The rubric does not replace case-specific reference criteria or the task-specific metrics of existing benchmarks. It has not yet been tested for inter-rater reliability, construct validity or clinical utility. Its immediate purpose is to make evaluation decisions explicit and open to scrutiny before empirical testing.

Self-supervised learning (SSL) by predicting in latent space, without generating the input data itself, learns highly abstract, useful representations. Intuitively, this success is often attributed to its ability to discard nuisance information that is irrelevant to prediction. However, this poses a conundrum: both stochastic variation in a prediction-relevant latent signal and true nuisance make observations partly unpredictable; how could they be distinguished? Surprisingly, we prove that common SSL methods can achieve exactly this, by implicitly instantiating a latent-variable model with stochastic dynamics and observation-private nuisance. We trace their ability to recover the stochastic signal to two complementary principles: Predictive mutual information maximization ensures that representations retain the information needed for prediction, while latent distribution matching constrains how this information is encoded, thereby making the retained signal identifiable. We confirm this identifiability result in simulations for Gaussian predictors, which recover the true signal up to an affine transformation even in dynamic, nuisance-laden environments.

Speech self-supervised learning aims to learn general-purpose representations for downstream speech tasks. However, current approaches rely on complex, carefully designed prediction targets. We challenge this necessity with GLaS-JEPA, a framework that directly predicts the current encoder's continuous representations at masked positions, without contrastive learning, discrete targets, or separate EMA target encoders. We prevent representation collapse using SIGReg representation-space regularization, eliminating the need for engineered target-generation mechanisms. Pretrained on 960 hours of LibriSpeech, our 57M-parameter model achieves a 6.89% WER on frozen-encoder SUPERB ASR and a 25.87% CER on slot filling, outperforming the best non-distilled sub-90M baselines by 43.1% and 22.0%, respectively. These results demonstrate that highly competitive speech representations can emerge from a radically simplified training recipe.

Many recommender services repeatedly encounter cold-start cohorts, where new users arrive with little or no interaction history. This creates two challenges: learning user preferences quickly from limited feedback and sustaining useful recommendations when each user has a finite catalog that can become repetitive or depleted over time. We propose CohortMix-TS, a warm-started mixture bandit that learns latent user groups from earlier cohorts and uses available metadata to construct group-informed priors for new users. Starting from these fixed priors, the model personalizes independently as feedback from each user becomes available. Session slates combine Thompson sampling with diversity and inventory-depletion controls. We evaluate CohortMix-TS through simulation, semi-synthetic experiments, and a 25-day randomized in-the-wild deployment with 713 registered participants in a Campus Games quiz application. Our evaluations show that cross-cohort transfer improves early recommendation quality and user-level regret, while inventory-aware slate construction helps prevent premature exhaustion of preferred items. In the field deployment, treatment users also showed a larger early-to-late change in correctness than users receiving random recommendations. Together, these results show how warm-start transfer and inventory-aware recommendations can support personalization for short-lived, repeatedly cold-starting cohorts.

Vision-language-action and world-action models have demonstrated impressive capabilities in robotics, yet generalization to unseen tasks remains challenging. More recently, general-purpose multimodal agents have shown great potential for zero-shot robotic task solving. However, they often incur high execution costs by reasoning and exploring the physical world from scratch. To reduce these costs, we introduce RoboSkill, a framework that connects skill acquisition and reuse through an Explore, Execute, Evolve loop. Within this loop, the agent explores to gather task-relevant information, executes tasks while adapting to feedback, and evolves its skill library based on execution records. It then reuses these skills to guide exploration and execution in the next cycle, closing the loop. To improve loop efficiency, we complement vision with tactile feedback to reduce uncertainty during physical interaction. We further augment textual guidance with reusable code to reduce reasoning overhead during skill reuse. On LIBERO-10, RoboSkill improves first-episode success rates by 12.5--25.0 percentage points and reduces average runtime by 7.6--72.4% across four agents. On real robots, it improves success rates by 8.3 percentage points and reduces average runtime for successful trials by at least 14.4%.

Empathetic spoken dialogue requires models to use both what is said and how it is said to decide how to respond. Explicit CoT can improve paralinguistic perception and make acoustic cues more explicit in replies, yet does not ensure their effective use in response planning. We call this mismatch the perception-reasoning gap. In addition, CoT may not fully capture acoustic cues in words, and generating it adds inference latency. To address these limitations, we introduce LoopSLM, which builds on looped Transformers for latent reasoning, reusing a decoder block to refine hidden states with acoustic grounding at every pass. Its two-stage training further narrows the perception-reasoning gap by separating learning to reason from learning to respond, enabling direct inference without CoT. On EchoMind, LoopSLM improves paralinguistic understanding, reasoning, and reply quality over Qwen2.5-Omni-7B. Against the CoT-SFT baseline, LoopSLM gains over 20 points in reasoning accuracy while generating 64.5% fewer tokens at half the latency. It also outperforms Qwen3-Omni-Thinking on most empathetic reply metrics with 34x lower latency. Despite training only on dialogue data, LoopSLM improves accuracy on general audio benchmarks.

Electronic invoices are replacing paper invoices worldwide, but today's centralized architectures leave three problems unsolved on the consumption side: an invoice can be submitted for reimbursement repeatedly, authenticity is difficult for recipients to verify, and data is siloed at a central authority that forms both a performance bottleneck and a single point of failure. This paper presents the design, formal analysis, and implementation of a complete blockchain-based electronic invoice system on Ethereum. We formalize the invoice lifecycle as a guarded labeled transition system and prove, under standard cryptographic and consensus assumptions, that the system guarantees: (i) reimbursement uniqueness--an invoice is reimbursed at most once, even across mutually distrusting organizations; (ii) face integrity--any verified invoice matches the recorded one unless keccak256 second-preimage resistance is broken; and (iii) authorization soundness for every lifecycle operation. The core invariants are machine-checked using Solidity SMTChecker, proving inductive validity across all reachable transaction sequences. The architecture models each invoice as a non-fungible, non-tradable token whose state transitions through five guarded subsystems, employing a lock-based protocol that makes duplicate reimbursement unrepresentable rather than merely detectable. We implement the design as a Solidity 0.8 contract with a four-role web application and evaluate it on a private Ethereum network: issuing costs 646,773 gas, full reimbursement costs under 135,000 gas, all operations run in O(1) time, and a single node sustains 137 issuances/s. Finally, the verified contract serves as a safety envelope for LLM-based reimbursement agents, provably rejecting unsafe actions (duplicate, over-limit, or forged-receipt claims) even when the agent's internal policy fails. All code and benchmarks are open-source.

Assigning credit to intermediate steps remains a central challenge in training Large Language Models (LLMs) on multi-step reasoning tasks with sparse terminal rewards, and actor-critic methods such as PPO address this by learning value functions to construct token-level advantages. Their effectiveness, however, hinges on reliable value estimation, a difficult task requiring the critic to both assess progress toward a correct solution and anticipate an evolving policy's future behavior; errors in either can compromise credit assignment and destabilize online training. In this paper, we revisit the standard state-only formulation of value estimation and propose $\pi$PPO, a self-privileged actor-critic framework. By reusing verified same-prompt rollouts as contrastive evidence, $\pi$PPO helps the critic assess intermediate reasoning against successful and failed attempts, while preserving standard policy optimization and the deployment interface. Experiments show that $\pi$PPO consistently improves value-estimation quality by a substantial margin and outperforms representative actor-critic and critic-free RLVR baselines on challenging mathematical reasoning benchmarks, while remaining effective even when paired with substantially smaller asymmetric critics.

Influence functions estimate how individual training examples affect the behavior of large language models (LLMs). Analyzing how training data influence different behaviors of an LLM involves repeated influence computation. Reusing stored training gradients reduces the computational cost, but storing full gradients is prohibitively expensive at LLM scale. We study how to compress these gradients while preserving influence estimates for future queries that are unknown at storage time. Through a worst-case analysis, we characterize the optimal fixed-dimensional linear representation and propose eigenbasis-corrected one-bit gradient projection (EOGP) to approximate it at scale. Specifically, EOGP uses EK-FAC to reduce gradient dimensionality, then applies PCA within the retained subspace to learn compression directions from the training gradients. We then apply one-bit quantization to the resulting coordinates, allowing more coordinates to be retained within a fixed storage budget. On GPT-2, EOGP predicts retraining outcomes more accurately than the evaluated compression baselines while using one-sixteenth of their per-example storage. On OLMo 2 SFT models from 1B to 32B parameters, EOGP remains competitive with the baselines allocated over 100 times as much storage per example.

Scientific software is increasingly required to process larger datasets while maintaining acceptable execution times. Software optimization traditionally requires substantial expertise in programming, algorithms, and numerical methods. Recent advances in large language models (LLMs) offer the possibility of automating much of this process. We investigate whether LLM-based agents can autonomously achieve substantial performance improvements in scientific software, including mature implementations that have already been extensively optimized by human developers. We tasked an LLM-based agent with optimizing software for three computational problems: t-SNE, single-sample gene set enrichment analysis (ssGSEA), and graphlet counting. Humans defined the scope, correctness criteria, and a verification mechanism, after which the agent worked autonomously, in some cases for several hours. Code maintainers reviewed each resulting implementation and verified its correctness. The optimized implementations were faster in all tested configurations, by up to two orders of magnitude over the fastest existing tools. The improvements included low-level code optimizations, mathematical reformulations, and an entirely new algorithm for graphlet counting. Software optimization can increasingly be delegated to autonomous agents, with the human role shifting from implementing optimizations to deciding which software to optimize, defining objectives, providing verification mechanisms, and ensuring the correctness of the final software. For well-scoped, verifiable problems, we argue that manual software optimization may be a thing of the past.

Agent harness bugs exhibit unique characteristics and remain challenging for state-of-the-art software agents to repair. Progress in this area is further hindered by existing benchmarks, which contain only a small and fixed number of executable harness bugs while requiring hundreds of human hours to construct. This work presents AgentBug-Smith, an automated harness bug reproduction approach that continuously discovers and reproduces real-world harness bugs from open-source agentic systems. Across different backbone LLMs, AgentBug-Smith consistently outperforms existing bug reproduction techniques designed for general software, achieving 10.67% - 27.56% higher success rates of reproducing harness bugs. By applying AgentBug-Smith to open-source agentic systems in the wild, we construct Live-Harness-Bench, a live and extensible benchmark that currently contains 200 reproducible harness bugs. We further demonstrate the utility of Live-Harness-Bench through two downstream applications. First, we use Live-Harness-Bench as the evaluation benchmark to systematically evaluate state-of-the-art software agents, revealing their limited capabilities in repairing real-world harness bugs. Second, we use Live-Harness-Bench as a knowledge base of real-world harness bug fixes, from which reusable repair skills can be distilled to improve existing software agents, increasing their harness-bug repair rates by 6.32%. Together, AgentBug-Smith and Live-Harness-Bench establish a scalable foundation for continuously evaluating and improving software agents on harness bug repair, turning real-world agent failures into executable evaluation instances and reusable knowledge for harness improvement, thus contributing to the ultimate goal of recursively self-improving agents.

Reinforcement Learning with Verifiable Rewards (RLVR) methods such as GRPO rely on successful self-generated trajectories, but finite rollout budgets can produce all-fail groups with no reward-based policy-gradient signal. While additional rollouts improve the chance of success at higher cost, successful trajectories missing from one model's rollouts may already have been discovered by another. Indeed, we observe that heterogeneous models often succeed on complementary prompts, creating opportunities for mutual learning without a designated stronger teacher. To exploit this complementarity, we propose GRAFT (Gated Replacement of Answer-Failed groups with peer Trajectories), an off-policy-aware framework that replaces all-fail groups with informative peer groups. GRAFT transfers both successful and unsuccessful peer responses with peer-computed advantages, while controlling cross-model mismatch through sequence-level compatibility weighting and token-level importance ratio clipping. Across three heterogeneous model pairs and five mathematical reasoning benchmarks, GRAFT consistently improves both models over GRPO with the same per-model rollout budget, gaining 2.1 points on average and up to 4.5 points in model-level average performance. Stored peer trajectories preserve most of the gains, improving over GRPO by 1.8 points on average without simultaneous co-training.

Average performance on routine driving benchmarks does not establish planner reliability under rare, safety-critical hazards. We proposed ExceptionDrive, a counterfactual planning benchmark that uses VLM-assisted screening, localized multi-view editing, and quality auditing to insert hazards into real nuScenes scenes while preserving their context. Its 21 tasks span six safety families and define hazard or conflict regions, local safety constraints, and acceptable responses. Because hazard insertion can invalidate the recorded human trajectory, our reference-free protocol evaluates edited predictions using Unsafe Rate (UR), Hazard Clearance Compliance (HCC), Hazard Proximity Response (HPR), and Counterfactual Trajectory Shift (CTS), which measure core-region intrusion, clearance compliance, clearance relative to a prescribed margin, and counterfactual trajectory change. Seven representative planners frequently intrude into hazard regions or provide insufficient clearance. We also develop a Reminder Agent that, without sample-specific task labels, converts visual evidence and the shared taxonomy into structured records of hazard presence, type, and a recommended high-level strategy. The agent neither predicts trajectories nor controls the vehicle; its records guide a VLM-based decision agent. In zero-shot experiments, the reminders improve strategy accuracy and reduce under-warning.

Flocking and schooling are thought to have evolved partly as defences against predation, but how prey should balance social and escape tendencies may depend on the predator's hunting strategy. We extend the predator-prey boids model of Ojo et al. (2023), itself based on Reynolds' boids, by combining six prey movement tendencies (alignment, cohesion, separation, dodge, repel and wiggle) into a single weighted acceleration update, and by reformulating wiggle as a sinusoidal manoeuvre. We then use an evolutionary strategy to optimise the six behaviour coefficients for collective prey survival against four predator hunting strategies: attack-centroid, attack-nearest, attack-random and attack-peripheral. Across five independent trials per strategy, coefficients converged within trials and mean fitness remained stable or increased, although trials often settled in different local optima. Prey survival was highest under attack-centroid and lowest under attack-nearest, in line with our hypotheses. Against attack-centroid, prey evolved individualistic predator avoidance with high escape coefficients, whereas against the other three strategies they largely kept their flock formation. Across all strategies, evolution favoured a low repel coefficient and relatively high dodge and wiggle coefficients. Our results suggest that optimal anti-predator behaviour depends on the interplay between escape tendencies and the predator's hunting strategy.

Click-Through Rate prediction, a core task in recommendation and advertising systems, relies on modeling interactions among sparse categorical features. Explicit cross networks are a central paradigm for CTR prediction, and recent progress has largely come from increasing the interaction capacity of a single predictor through deeper cross networks and more expressive cross operators. We revisit whether continually increasing interaction capacity remains the most effective way to improve predictive performance, and find that its benefits quickly exhibit diminishing returns even as capacity continues to grow. This motivates a complementary scaling direction that we call estimator scaling, where additional resources are used to incorporate multiple related estimators rather than only enlarging a single predictor. Through theoretical analysis, we show that the gains from estimator scaling are governed by the amount of non-shared predictive variation available across estimators. However, exploiting this variation naively can be expensive: independently trained models provide substantial estimator diversity but require deployment cost to grow with ensemble size. This motivates a parameter-efficient realization of estimator scaling that can incorporate diversity from multiple estimator sources without maintaining multiple full models. Building on this view, we introduce RECursive Averaged Predictor (RECAP), a parameter-efficient recursive CTR model that operationalizes estimator scaling at three levels: distillation across independently trained models, exponential moving averaging over training trajectories, and aggregation over inference-time routes within a weight-shared recursive backbone. Experiments across multiple benchmarks establish new state-of-the-art predictive performance on standard benchmarks, while placing the RECAP on a favorable performance-parameter Pareto frontier.

Scientific queries are often brief, while relevant papers use specialized vocabulary. Generated query expansion can bridge this mismatch, but earlier work suggests that its value shrinks as the underlying retriever becomes stronger. We test the four generated formats of term lists, a pseudo-document, multiple pseudo-references, and corpus-steered text all together with SPLADE-v3 on NFCorpus, TREC-COVID, and SciDocs. Every condition searches the same frozen document index and follows the same query-side integration rule and 256-dimension budget, isolating the effect of the added content. All twelve method-collection comparisons improve aggregate nDCG@10, with best relative gains of 4.81%, 8.92%, and 9.47%. Eleven remain significant after Holm correction. The gain persists in 103 of 114 interpolation settings, including every setting that assigns at least 30% of the mixture weight to the original query. Shuffled-text and non-contextual lexical-bag controls also remain above baseline in all 24 aggregate comparisons, showing that the added vocabulary carries most of the benefit. A corpus-induced typed concept graph, by contrast, produces no consistent gain, and its relation, depth, validation, random, and gating controls do not rescue it. Generated vocabulary can therefore complement a strong learned sparse retriever, provided that the original query remains strongly represented.

Fine-tuning large language models on narrow, misaligned tasks can undo their post-training alignment and induce novel misaligned behaviors -- a phenomenon known as \emph{emergent misalignment} (EM). EM has been linked to persona-like representations, where fine-tuning might reduce loss by amplifying a harmful or 'evil' persona. It remains unclear which properties of the training data drive this effect: whether all harmful examples contribute approximately equally to misalignment and whether different models are equally affected by the same fine-tuning examples. In this work, we use training data attribution to quantitatively estimate how much each harmful example contributes to EM. We benchmark the quality of the attribution via retraining -- a sound attribution score should enable us to enhance or attenuate EM by filtering data on that score. Score-based filtering can substantially enhance or attenuate EM; we find that both data-attribution scores and a black-box harmfulness score can identify consequential examples. All models we test become misaligned when trained on the same dataset, and influence scores perform best when filtering data from the same model that computed them. We find cross-model generalization of influence scores from scores derived from the three model families we tested, but this generalization does not recover same model filtering performance.

Production machine learning (ML) stacks often split graph compilation and kernel execution across different layers and languages, making backend behavior, deployment guarantees, and performance fallbacks hard to reason about end-to-end. RLX addresses this gap with a single Rust codebase that combines compiler and runtime roles around one primitive-level, three-level intermediate representation (IR), plus a transparent dispatch contract that resolves each operator to native, common-IR, or rewritten lowering and fails compilation when legalization is not possible. The same IR targets fourteen runtime devices (cpu, metal, mlx, ane, cuda, rocm, oneapi, tpu, hexagon, gpu, vulkan, opengl, directx, webgpu) and two specialty codegen paths (Cortex-M INT8 and FPGA), ingests safetensors, GGUF, ONNX, and rten formats, supports F16/BF16/F64/C64 and quantized INT4/INT8 flows with AMP/PTQ/QAT, and scales via tensor-/pipeline-parallel collectives over TCP and RDMA transports. Beyond neural workloads, RLX also extends to scientific/physics-style domains through sparse and dense linear algebra extensions (e.g., CSR LU/CG/matvec and LAPACK- backed factorizations) and 3D Gaussian splatting operators. We evaluate RLX against PyTorch, TensorFlow, JAX, candle, burn, tch, rten, MLX, CoreML, IREE, Glow, TensorRT, and tinygrad under identical input generation and p50 measurement methodology on one host. On all-MiniLM-L6-v2, RLX-Metal is fastest at every batch (e.g., 16.6 ms at batch 32 vs. PyTorch-MPS 26.7 ms). In the MNIST training table, RLX also has the top-throughput entry (graph-fused MLP: 946,487 img/s), above NumPy+BLAS (787,349 img/s), while retaining 100% top-1 parity on reference checks (e.g., Qwen3).

As Graph Neural Networks (GNNs) are widely deployed as Machine Learning-as-a-Service (MLaaS) APIs, model stealing attacks have emerged as a critical security threat. By querying a victim model's black-box API, an adversary can construct a functionally equivalent surrogate model, compromising proprietary intellectual property and downstream security. Existing GNN stealing attacks, however, rely on overly permissive assumptions, such as soft-label outputs, large query budgets, full-graph query access, and prior knowledge of victim backbones that rarely hold in real-world deployments. In this work, we formalize a strictly constrained black-box, hard-label and backbone-agnostic threat model for GNN stealing attacks under a tight query budget. Given these realistic restrictions, we identify four fundamental challenges: sparse local structures and isolated nodes that degrade victim label quality, insufficient supervision signals, systematic imbalance with incomplete class coverage, and backbone mismatch. To address these interlocking barriers, we propose Dagger, a novel two-phase decoupling-based attack framework. Specifically, in Phase 1, Dagger pre-trains a surrogate using decoupled information propagation to preserve structural context over sparse local subgraphs while handling isolated nodes, combined with manifold-level node mixup to synthesize continuous supervision signals and smooth decision boundaries. In Phase 2, Dagger freezes the encoder and fine-tunes the classifier head via class-balanced sampling paired with logit adjustment to rectify severe query imbalance without requiring extra victim queries. Extensive experiments across four benchmark graphs and four GNN backbones demonstrate that Dagger consistently outperforms state-of-the-art GNN stealing attacks, achieving up to 18.16\% higher fidelity while only utilizing 12.23$\times$ fewer queries than the strongest baseline.

Masked diffusion models (MDMs) generate text by unmasking several tokens per step, but they are trained and sampled under different conditions. The model is trained on randomly masked sequences, whereas inference follows a trajectory shaped by the model's own predictions. Additionally, each step has no access to what the previous one computed. Recent methods narrow these limitations from separate angles, leaving open how these choices interact. We introduce PUMBA, a unified framework for trajectory-aware training that trains the denoiser on consecutive steps of policy-induced trajectories, passes information between steps, and optimizes them jointly by backpropagation through time. A controlled study of this design space shows that i) exact train--inference alignment fails due to local overfitting, whereas a looser alignment still brings training masks closer to those seen at inference; ii) passing continuous information outperforms discrete gradient estimators through the commitment at each step; and iii) performance improves as backpropagation through time spans more steps, which we support theoretically. Combined, these components match the best checkpoint of a same-size autoregressive model. Building on these findings, we scale PUMBA to supervised fine-tuning of LLaDA-8B, where it improves the trade-off between performance and number of function evaluations (NFEs) in both full-canvas and block diffusion generation. At matched performance, it needs up to 22% fewer NFEs than standard fine-tuning with twice the budget in full-canvas generation, and up to 26% fewer than standard fine-tuning for the same number of steps in block diffusion.

The BITEM team entered both subtasks of the NTCIR-19 R2C2 task with a single agentic pipeline, in which a model searches, reads and records evidence over a movie corpus while an orchestrator holds the record and rules on what may be submitted. A claim is admitted only once an entailment cascade has checked it against the passage it cites, and an answer is released only once enough checked evidence stands behind it. Each question is run three or four times, every pass retrieving from a corpus stripped of what the earlier passes have already seen. The confidence filed with each answer is computed by the orchestrator from what the run leaves behind and is never asked of the model, which is offered no way to rate itself. The two retrieval runs placed 4th and 5th of 22, pooling the passes was worth 0.0709 nDCG@20, and the gain was largest on the multi-hop and post-processing-heavy questions, where the organisers rank the pooled run top of the field. Sixteen of the 25 answer runs were built on passages these two runs supplied, 12 of them filed by other teams. HMR rewards a system whose confidence is high where it answers right and low where it answers wrong. The pipeline reached an accuracy of 0.9219, 6th of 25, while the confidence filed with those answers gave an HMR of 0.4915, 13th. A few rules crafted over those same recorded signals, with no further model call and no further retrieval, raise that to an accuracy of 0.9375, 5th, and an HMR of 0.6985, 9th. Ranking on HMR alone can reward a system for answering wrongly with low confidence, so we propose accHMR, the accuracy multiplied by HMR, which reports the reward in proportion to the accuracy, and on which the revised rules would have scored 0.6549, 5th. For future work, fitting a model on the numbers the pipeline already produces, rather than writing such rules by hand, would be a real step forward.

Time series anomaly detection (TSAD) plays a crucial role in healthcare, finance, industrial monitoring, and other sectors. Within and between these settings, anomalies span vastly different temporal scales, from sub-second point spikes to multi-hour drift patterns. However, most existing TSAD methods commit to a single temporal granularity, and multi-scale designs either analyze different scales in isolation or are constrained to a predefined coarse-to-fine hierarchy, both failing to sufficiently capture multi-scale interactions. To resolve this limitation, we propose Multi-Scale Autoencoder with Cross-Scale Attention for TSAD (MSCAD), a simple yet powerful semi-supervised TSAD framework founded on parallel autoencoder branches corresponding to different patch sizes. A stack of symmetric bidirectional cross-scale attention blocks enables every pair of scales to exchange information before reconstruction without allowing any single scale to be privileged. On the comprehensive TSB-AD benchmark (40 datasets, 530 series), MSCAD achieves large performance gains against 50 baselines across multiple metrics, with VUS-PR of 0.57(+9.6%) on the univariate split and 0.47(+9.3%) on the multivariate split compared to the state-of-the-art.

We evaluate an auditable long-term memory system on LongMemEval-S. Its retrieval chain uses hybrid candidate retrieval, cross-encoder reranking, coverage-first packet compilation, and deterministic reasoning scaffolds; an LLM is used only as a replaceable final reader. The chain places all gold sessions in the candidate pool for 468/470 answerable questions and produces gold-complete packets for 462/470. With a Claude Opus reader called through an unpinned CLI alias, two 500-question passes score 479/500 and 475/500 under GPT-4o. The 72 answerable knowledge-update rows used a substantively modified scoring prompt whose effect under the official text has not been measured. The pair straddles Chronos High's published 478/500; differences in reader generation, scoring prompt, and possibly data version, plus within-system variance, establish neither superiority nor equivalence. A grok-4.6-high reader on the same packets scores 476/474, while a maximum-reasoning-effort agentic variant regresses to 461/465. The headline passes differ on eight verdict-flip rows. A second judge agrees with the headline judge on 493/500 rows (98.6%) in each pass and scores both passes 472/500; the official judge also flips three verdicts when re-scoring byte-identical pass-1 answers. Negative controls rejected a verifier that repaired three wrong drafts but broke eleven correct drafts. All components were developed on the same 500 questions, with no held-out evaluation or independent human adjudication; retrieval and scaffold method sources and transcript-derived audits are held; and the headline reader received extra operator context, its complete requests were not retained, and MCP tool availability is unresolved. We release materialized packets, scaffolds, reader outputs, judge verdicts, and controls for inspection and re-scoring.

On-policy distillation (OPD) trains a student on its own generated responses using dense, token-level supervision from a stronger teacher. Vanilla OPD treats all teacher signals equally, assuming that the teacher's supervision is equally important for every token. However, teacher signals at different tokens may have very different effects on the student's performance: some correct important reasoning errors, while others have little effect on the final answer. Motivated by this observation, we introduce Dr. OPD (OPD Done Right), which defines the optimal weighted OPD to maximize the student's performance. We formulate Dr. OPD as a bilevel optimization problem in which the student learns from weighted teacher supervision, while the weights are selected to maximize the expected reward of the resulting student. To solve Dr. OPD, we develop an efficient iterative solver that updates the token weights and student policy alternatively. At each round, it updates weights in closed form and then takes one gradient step on the resulting weighted OPD objective. Under regularity conditions, we show that this weighted update achieves a higher expected reward than a vanilla OPD update. Empirically, across strong-to-weak and same-size distillation on math and code, Dr. OPD consistently outperforms all evaluated baselines. In particular, in the strong-to-weak distillation setting, Dr. OPD improves average math performance by $9.7$ points over vanilla OPD, and enables the smaller student to surpass its larger teacher.

Understanding gender biases in large language models (LLMs) is increasingly important as these systems become embedded in decision-support tools with real consequences. Prior research has focused only on a small set of models, leaving open the extent to which gender biases are common and heterogeneous across LLMs. We address this gap across ten models released between April 2025 and June 2026, spanning nine vendors, using two paradigms: gender attribution to stereotyped phrases (Study 1) and moral judgment of abuse or torture against a woman or a man to prevent a catastrophic outcome (Study 2). In Study 1, two of ten models attributed masculine-stereotyped phrases to female writers more often than the reverse, while three models showed the opposite pattern. In Study 2, several models converged on a male-disadvantaging asymmetry that was directionally consistent with a documented human tendency to protect female targets from harm, though the specific conditions under which this asymmetry emerged varied by model; three other models, by contrast, showed no variation across conditions. These results indicate that gender-related biases are common in LLMs. Their direction and magnitude, however, are highly heterogeneous, to the point that some models behave in diametrically opposite ways to others. Bias auditing should therefore be treated as an ongoing, multi-vendor process, rather than a one-time assessment.

Training agents to follow arbitrary instructions is an important goal of multi-task reinforcement learning (RL). Linear temporal logic (LTL) provides a precise and structured formalism for specifying instructions to agents, and has been successfully adopted for training generalist multi-task policies. However, differences in implementations, task distributions, and evaluation protocols make existing methods difficult to compare, while high computational costs limit the scale and statistical reliability of experiments. We introduce Jaxolotl, a unified high-performance benchmark suite for multi-task LTL-RL to address these concerns. Jaxolotl provides a modular, end-to-end JAX implementation of six representative algorithms and four environments, together with newly curated task suites and a standardised, statistically robust evaluation protocol. By precompiling symbolic task representations into static arrays, Jaxolotl enables fully JIT-compiled training and evaluation, achieving end-to-end speedups of up to $220\times$ and supporting controlled comparisons at substantially greater experimental scale. We use this framework to systematically evaluate existing approaches, revealing complementary strengths and limitations: general methods capable of non-myopic reasoning struggle as the number of propositions grows, while methods with stronger scaling rely on environment-specific assumptions and suffer from myopia.

Ship structural vibrations contribute to noise, fatigue, and equipment damage, while dynamic-compliance topology optimization can produce pathological designs near resonance. This study extends neural-reparameterized topology optimization using a convolutional Kolmogorov-Arnold network (KATO) to forced-vibration design with active input power (AIP) as the objective. Applications include a 100 Hz engine-supporting deck panel and an 18 Hz thruster foundation frame. Helmholtz PDE filtering and Heaviside projection control feature sizes and manufacturing tolerance. Across both deck families, all eight optimized layouts reduce AIP relative to size-optimized references and, after finite-depth extrusion, also achieve lower static compliance. For unrestricted, manufacturing-aware, and stress-aware frame variants, KATO matches GCMMA in AIP within 0.5 dB while yielding 22-36x lower static compliance after matched-volume binary re-analysis. In a near-resonant 300 Hz case, both methods reduce initial AIP by more than 32 dB; KATO maintains a connected design, achieves 59x lower binary static compliance, and reduces maximum AIP over 1-500 Hz by 2.7 dB. KATO runs 6.4-10.4x faster than GCMMA for the implemented stress-aware formulations. The results demonstrate neural AIP-driven topology optimization as an efficient approach for designing connected, feature-size-controlled ship structures with improved forced-vibration performance.

Chain-of-thought traces are widely read as records of how models reach their answers, informing debugging, agent auditing, and claims about reasoning. Testing this interpretation is difficult because natural-language thinking traces are rarely mechanically verifiable. We revisit it in iGSM, a synthetic grade-school mathematics benchmark designed to study thinking traces and used to support claims of learned reasoning and planning. Crucially, iGSM exposes the exact quantities and dependencies that a correct solution should use, allowing generated traces to be checked programmatically step by step and enabling us to test whether correct answers are reliably accompanied by valid traces. We first evaluate models trained exclusively on valid, minimal traces. Answer correctness and trace validity nearly coincide in distribution but decouple out of distribution: on the hardest instances, 31.6% of correct answers have invalid traces, over half of which pass all syntactic and arithmetic checks but fail semantic dependency checks. We then intervene on trace supervision. Non-minimal training traces induce non-minimal outputs, while re-asking the same problem with a different query reveals computations inherited from the original query, weakening minimality as evidence of selective planning. Shuffling tokens in 10% of training trace sentences preserves near-clean accuracy even out of distribution despite no trace passing verification. Swapped training traces likewise retain high in-distribution accuracy. We discuss the implications of these findings for chain-of-thought monitoring and interpretation in the context of AI safety.

The attention operation is naively position invariant. However, positional information is fundamental to natural language, and therefore a variety of explicit position encodings have been developed in transformer-based models, such as rotary position encoding (RoPE). Although explicit position encodings have long been assumed to be required, recent methods that interleave local mixing layers, such as sliding window attention (SWA) and gated linear attention, while not encoding position (NoPE) in global attention layers has recently been shown to be successful at scale. How and why this approach works is not well-understood. In this paper, we develop an explanation of how hybrid models of this sort can implicitly encode position at global NoPE layers. Supported by both theoretical and empirical evidence, our central argument is that SWA and gated linear attention induce a recency bias in the residual stream that propagates to, and is selected by, the global attention logits. Moreover, in contrast to the implicit position encodings found in models with only global NoPE attention, in which positional information arises solely from the causal mask, the recency bias in hybrid models can be maintained across long sequences. In addition to deepening our understanding of how hybrid models encode position, these findings may provide insights for how to encode position in a way that can extrapolate to longer sequence lengths indefinitely.

Recent LLMs increasingly adopt hybrid designs that replace standard attention with linear attention, such as Gated DeltaNet (GDN) and Kimi Delta Attention (KDA). Although they compress the context into a fixed-size recurrent state and substantially reduce the cost of long-context processing, repeatedly reading and updating that state remains a major inference bottleneck. Quantization offers a natural way to reduce this cost, but can significantly degrade model quality, due to the accumulation of rounding errors and the presence of outlier rows and columns in the state. To address these challenges, we propose LeapQuant, a training-free method that achieves near-lossless performance under 8-bit recurrent-state quantization. First, to mitigate error accumulation, we propose per-window quantization, which leaps over a window of tokens and quantizes the state only once at its end. Within a window, outputs are computed from the fixed low-bit state together with high-precision buffered updates. Second, to reduce the error introduced by each quantization, LeapQuant retains the state's largest outliers as a few high-precision Compensator Tokens, which share the update path of real tokens. We then smooth the remaining residual before quantization to further reduce the error. Comprehensive experiments across the Qwen, Kimi, and GLM model families show that LeapQuant substantially reduces memory and compute costs during inference. With accuracy comparable to the FP32 baseline, it achieves average speedups of 2.05--3.70$\times$ at the kernel level and 1.47$\times$ for end-to-end inference on NVIDIA B200, RTX PRO 6000, and RTX 5090 GPUs.

Linear attention replaces growing KV caches with fixed-size recurrent states, yet these persistent states can become a substantial memory bottleneck under concurrent serving. Directly quantizing recurrent states to low precision often leads to severe accuracy degradation, as quantization errors propagate through successive state updates. We discover that the impact of these errors depends on two complementary dimensions: temporally, errors in long-lived memory can persist across many decoding steps; spatially, errors in different key rows affect model outputs differently, while state magnitudes vary substantially along both rows and columns. Motivated by these observations, we propose STEPQuant, a spatial-temporal post-training quantization framework for Delta-rule recurrent states. STEPQuant allocates precision according to error magnitude and memory lifetime, and jointly fits key-row and value-column scales based on state distributions and key-row impact on output error. Experiments on Qwen3.8-27B and Kimi-Linear-48B-A3B-Instruct across both long- and short-generation benchmarks show that STEPQuant closely matches FP32-state accuracy under a nominal 6-bit budget and outperforms uniform INT8 in its 4-bit configuration. Integrated into SGLang with optimized GPU kernels, 6-bit STEPQuant achieves over 5x recurrent-state compression and reduces total serving memory by up to 68.7%. Our code is available at https://github.com/Dreamer-Toby/STEPQuant.

Robots deployed in the physical world must be able to improve beyond their initial training as they encounter new situations and failures. For this improvement to scale across tasks, it must make effective use of experience without requiring human demonstration of each correction. Recent agentic systems offer a way to reduce this reliance on human effort by using foundation models to autonomously compose learned behaviors to complete tasks. Yet completing tasks this way does not itself teach a task policy to overcome its own failures; that requires turning these behaviors into learnable corrections for the policy. Our insight is that many such corrections are familiar short behaviors, or skills: they recur across tasks and describe actions that foundation models can reason about from a scene. We introduce skill-space shooting, which uses foundation model guidance to explore corrections through these reusable skills and turn successful trials into policy improvement. Real-world experiments show repeated improvement in policies acting autonomously, while skills can also be shared to reduce the teaching needed to improve on new tasks. By making reusable skills a source of corrective supervision, skill-space shooting enables scalable and generalizable policy improvement within and across tasks. Additional results and videos at https://skill-space-shooting.github.io.

Modeling scheduling problems with conditional time intervals and cumulative functions has become a common approach when using modern commercial constraint programming solvers. This paradigm enables the modeling of a wide range of scheduling problems, including those involving producers and consumers. However, it is unavailable in existing open-source solvers and practical implementation details remain undocumented. In this work, we present an implementation of this modeling approach using a single, generic global constraint called the Generalized Cumulative. We also introduce a novel timetabling filtering algorithm specifically designed to handle tasks defined on conditional time-intervals. Experimental results demonstrate that this approach, combined with the new filtering algorithm, performs competitively with existing solvers enabling the modeling of producer and consumer scheduling problems and effectively scales to large-scale problems.

Constraint Programming (CP) offers an intuitive, declarative framework for modeling Vehicle Routing Problems (VRP). While classical successor-based CP models can be adapted to handle optional visits or insertion-based heuristics, sequence variables provide a significantly more natural and elegant formulation for these requirements. Building upon our prior work that introduced the initial concept, the main contribution of this article is the complete semantic and operational formalization of sequence variables as a computational domain. Specifically, we formally define the sequence domain and its update operations, and detail the implementation and data structures required to integrate sequence variables into trail-based CP solvers. Furthermore, we introduce consistency levels for associated constraints on this domain alongside specialized global constraints tailored for routing problems. Finally, we demonstrate that sequence variables simplify problem modeling while achieving competitive computational performance on Pickup and Delivery Problems with and without Time Windows, the Dial-a-Ride Problem, and a Prize-Collecting Scheduling Problem.

Prediction markets allow users to trade on outcomes of real-world events, but are prone to fragmentation with overlapping questions, implicit equivalences, and hidden contradictions across markets. We present an agentic AI (AAI) pipeline that autonomously recovers cross-market structure from contract text before prices enter the analysis. The workflow first clusters markets into coherent topical groups using natural-language understanding over contract text and metadata, and then identifies contracts within each cluster, but from different event markets, that exhibit strong dependence or leader--follower relationships. We evaluate this system, along with a natural language inference (NLI) benchmark, on a large prediction market dataset from early 2026. Using resolved outcomes to evaluate identified relations, we find that AAI-identified relations are 62.8\% consistent with exchange-recorded settlements, whereas the NLI benchmark only achieves 40.6\% accuracy. Within clusters, the AAI output is sparse and also remarkably compatible as a signed graph with a frustration rate of 0.324\%. As an application, we show how discovered relations inform semantics-based trading strategies on prediction markets. One such strategy yields 14.12\% net ROI after fees in a two-month period in 2026. Overall, we demonstrate the potential for agentic AI as a structural discovery layer for prediction markets.

Low-rank adaptation (LoRA) approximates the update of a pretrained weight matrix using the product of two low-rank matrices. However, standard LoRA follows an explicit-rank paradigm, where increasing model capacity requires adding more rows or columns (i.e., basis vectors) to the low-rank matrices, leading to substantial parameter growth. In this paper, we find that these basis vectors exhibit significant parameter redundancy and can be compactly represented by lightweight nonlinear functions. Therefore, we propose Generative Low-Rank Adapter (GenLoRA), which replaces explicit basis vector storage with nonlinear basis vector generation. Specifically, GenLoRA maintains a latent vector for each low-rank matrix and employs a set of lightweight radial basis functions (RBFs) to synthesize the basis vectors. Each RBF requires far fewer parameters than an explicit basis vector, enabling higher parameter efficiency in GenLoRA. Extensive experiments across multiple datasets and architectures show that GenLoRA attains higher effective LoRA ranks under smaller parameter budgets, resulting in superior fine-tuning performance. The code is available at https://anonymous.4open.science/r/GenLoRA.

Flowchart-oriented dialogue (FOD) systems aim to guide users through multi-turn decision-making or operational procedures by following a domain-specific flowchart to achieve a task goal. In this work, we formalize flowchart reasoning in FOD as grounding user input to flowchart nodes at each dialogue turn while ensuring node transition is consistent with the correct flowchart path. Despite recent advances of LLMs in task-oriented dialogue systems, adapting them to FOD still faces two limitations: (1) LLMs lack an explicit mechanism to represent and reason over flowchart topology, and (2) they are prone to hallucinations, leading to unfaithful flowchart reasoning. To address these limitations, we propose FloCA, a zero-shot flowchart-oriented conversational agent. FloCA uses an LLM for intent understanding and response generation, while delegating flowchart reasoning to an external tool that performs topology-constrained graph execution, ensuring faithful and logically consistent node transitions across dialogue turns. We further introduce an evaluation framework with an LLM-based user simulator and five new metrics covering reasoning accuracy and interaction efficiency. Extensive experiments on FLODIAL and PFDial datasets highlight the bottlenecks of existing LLM/VLM-based methods and demonstrate the superiority of FloCA. The code and dataset are publicly available at https://github.com/Jinzi-Zou/FloCA-flowchart-reasoning.

LLM agents are increasingly expected to operate as general-purpose systems that resolve real-world user requests, yet their dynamic scaling behavior in realistic environments remains poorly understood. In this paper, we systematically investigate two principal test-time scaling axes of LLM agents: sequential scaling through extended interaction and parallel scaling through trajectory sampling. We first introduce a realistic benchmark that provides one unified framework for evaluating LLM agents across search, coding, reasoning, and tool-use domains, more faithfully reflecting the heterogeneity of real-world deployments. Evaluating ten leading LLM agents reveals substantial performance degradation when transitioning from domain-specific evaluations to this realistic setting. Building on this foundation, we progressively scale test-time compute along fine-grained increments to characterize the performance upper bound. We find that neither scaling axis can consistently yield meaningful gains from additional test-time compute in realistic environments, a phenomenon we attribute to two fundamental limitations: the scaling plateau that bottlenecks sequential scaling and the verification gap that undermines parallel scaling. Code is publicly available at https://github.com/cxcscmu/General-AgentBench.

This study proposes CIRCLE, a six-stage, lifecycle-based framework to bridge the reality gap between model-centric performance metrics and AI system outcomes in deployment. Current approaches such as MLOps frameworks and AI model benchmarks offer detailed insights into system stability and model capabilities, but they do not provide decision makers outside the AI stack with systematic evidence of how these systems actually behave in real world contexts or affect their organizations over time. CIRCLE operationalizes the Validation phase of TEVV (Test, Evaluation, Verification, and Validation) by translating priorities of stakeholders outside the stack into measurable signals. Unlike participatory design which often remains localized, or algorithmic audits which are often retrospective, CIRCLE provides a structured, prospective protocol for linking context sensitive qualitative insights to scalable quantitative metrics. By integrating methods such as field testing, red teaming, and longitudinal studies into a coordinated pipeline, CIRCLE produces systematic knowledge; evidence that is comparable across sites yet sensitive to local context. This can enable governance based on materialized downstream effects rather than theoretical capabilities.

Temporal knowledge graph (TKG) extrapolation is an important task that aims to predict future facts through historical interaction information within KG snapshots. A key challenge for most existing TKG extrapolation models is handling entities with sparse historical interaction. The ontological knowledge is beneficial for alleviating this sparsity issue by enabling these entities to inherit behavioral patterns from other entities with the same concept, which is ignored by previous studies. In this paper, we propose a novel encoder-decoder framework OntoTKGE that leverages the ontological knowledge from the ontology-view KG (i.e., a KG modeling hierarchical relations among abstract concepts as well as the connections between concepts and entities) to guide the TKG extrapolation model's learning process through the effective integration of the ontological and temporal knowledge, thereby enhancing entity embeddings. OntoTKGE is flexible enough to adapt to many TKG extrapolation models. Extensive experiments on five data sets demonstrate that OntoTKGE not only significantly improves the performance of many TKG extrapolation models but also surpasses many state-of-the-art(SOTA) baseline methods.

Recent work has used LLM agents to reproduce empirical social science results with access to both the data and code. We broaden this scope by asking: Can they reproduce results given only a paper's methods description and original data? We develop an agentic reproduction system that extracts structured methods descriptions from papers, runs reimplementations under strict information isolation -- agents never see the original code, results, or paper -- and enables deterministic, cell-level comparison of reproduced outputs to the original results. An error attribution step traces discrepancies through the system chain to identify root causes. Evaluating four agent scaffolds and four LLMs on 48 papers with human-verified reproducibility, we find that agents can largely recover published results, but performance varies substantially between models, scaffolds, and papers. Root cause analysis reveals that failures stem both from agent errors and from underspecification in the papers themselves.

Time-series data are inherently multiscale, spanning diverse temporal granularities from coarse trends to fine-scale dynamics. However, existing time-series generative models provide limited control over the temporal granularity of both inputs and outputs, restricting their ability to condition on user-provided coarse sketches and generate samples at a desired target granularity. To address this, we introduce TimeTok, a unified framework for Granularity-Controllable Time-Series Generation (GC-TSG), which generates time series at any target granularity from any coarser input (e.g., rough sketches) or without conditioning. At the core of TimeTok is a hierarchical tokenization strategy that maps time series into an ordered sequence of tokens, from coarse to fine temporal granularity. Our autoregressive generation process operates across these granularity levels, producing token blocks that are decoded back into continuous time series. This design naturally enables GC-TSG within a single framework, where controlling the number of token blocks provides explicit control over output detail. Experiments show that TimeTok excels at GC-TSG tasks while achieving state-of-the-art performance in standard generation. Furthermore, we showcase TimeTok's potential as a foundational tokenizer by training on multiple datasets with heterogeneous temporal granularities, verifying strong transferability that consistently outperforms models trained on individual datasets. To our knowledge, this is the first unified framework that covers the full generative spectrum for time series, offering a valuable foundation for models that benefit from diverse temporal granularities.

Natural-language instance navigation becomes challenging when the initial user request does not uniquely specify the target instance. A practical agent should reduce the user's burden by actively asking only the information needed to distinguish the target from similar distractors, rather than requiring a detailed description upfront. Existing approaches often fall short of this goal by mistaking distractors that strongly match the accumulated information about the target provided by the user. As a result, despite the dialogue, the agent may still fail to distinguish the target from distractors, leading to premature decisions and lengthy user responses. We propose Proactive Instance Navigation with Comparative Judgment (ProCompNav), a two-stage framework that first constructs a candidate pool and then identifies the target through Recursive Comparative Judgment (RCJ). RCJ iteratively narrows the pool by selecting an attribute-value pair that divides the candidates, asking the user a binary question, and removing inconsistent candidates, without requiring an attribute unique to the target. On CoIN-Bench, ProCompNav outperforms the evaluated baselines in Success Rate while substantially reducing Response Length. On the non-interactive TextNav benchmark, ProCompNav achieves the highest Success Rate. Two human studies further show that participants prefer ProCompNav's interaction strategies.

Reliable human-machine discrimination is becoming increasingly important as Large Language Models and autonomous agents are deployed in online settings. Existing approaches evaluate whether a system can produce responses indistinguishable from those of a human. This approach follows the focus on the output of a machine, as suggested by Alan Turing. Cognitive science provides an alternative approach: considering the process by which that behavior is produced. To evaluate whether processes can reliably distinguish humans from machines, we introduce a process-based framework, the Process Turing Test, and evaluate it across a battery of cognitive tasks spanning decision-making, working memory, and planning. These tasks, such as mental rotation and sequence prediction, yield process-level measures complementing conventional measures of overall task performance. We also include multiple CAPTCHA tasks in the battery. Across the battery, process-level features provide substantially stronger discriminative signal than performance metrics alone, reliably distinguishing humans from agents even when task performance is matched (process-based classifier AUC = 0.88). We also conducted a controlled red-teaming study comparing off-the-shelf frontier agents (Claude Sonnet 4.5, GPT-5, Gemini 2.5 Pro), Centaur (LLM fine-tuned on 10.7M human decisions), and two task-specific fine-tuning methods: action-level supervised fine-tuning (A-SFT) and process-level fine-tuning (P-SFT), which directly optimizes process features. We find that broad fine-tuning on human choices makes task processes more human-like relative to off-the-shelf frontier agents, and task-specific P-SFT further improves human-like behavioral mimicry, though this advantage largely disappears under cross-task transfer. These results highlight process specification as a central bottleneck in achieving human-like cognitive processes in machines.

Multi-agent debate (MAD) systems increasingly rely on shared memory to support long-horizon reasoning, but this convenience opens a critical vulnerability: a single corrupted entry can contaminate the downstream memory-augmented reasoning, and debate alone fails to filter such errors. Existing safeguards filter entries via heuristics or LLM-based validation, yet they rely on AI judgments that share the same failure modes and overlook the cross-agent dynamics of MAD. We address this gap by formulating memory updating in MAD as a zero-trust memory game, in which no agent is assumed reliable and the game's equilibrium motivates a principled objective for calibrating memory influence. Guided by this objective, we propose EquiMem, an inference-time calibration mechanism that evaluates each update against the shared memory state and the memory already being used by other agents, using their existing retrieval queries and traversal paths as evidence of calibration. EquiMem instantiates this calibration for both embedding- and graph-based memory, and across diverse benchmarks, MAD frameworks, and memory architectures, it consistently outperforms existing safeguards, remains robust under adversarial agents, and incurs negligible inference overhead.

Language agents are increasingly deployed in professional workflows, yet tutoring remains a high-stakes capability that existing evaluations only partially capture. Effective tutor agents require more than producing correct answers or executing accurate tool calls: they must infer a warranted teaching decision from evidence, adapt support as learner state changes, and carry an instructor's request through a learning-management system (LMS) to a completed, verified intervention. We introduce TeachArena, a source-grounded benchmark that jointly evaluates three complementary surfaces of teaching work: professional pedagogical judgment, situated multi-turn tutoring, and end-to-end LMS teaching workflows. Its 354 audited tasks are each built around a pedagogical insight, grounded in evidence, and evaluated with matched verifiers over observable turn-level responses, tutoring trajectories, and persistent artifacts or environment states. Across a comprehensive evaluation of frontier models, our findings reveal that current models are generally capable of bounded pedagogical judgment, but still fall short of professional teaching standards in situated tutoring and end-to-end teaching-workflow execution. By unifying teacher judgment, adaptive tutoring, and institutional action in one auditable benchmark, TEACHARENA provides a measurement foundation for developing tutor agents that can support realistic teaching work.

Large language models (LLMs) provide a flexible foundation for multi-agent systems, but their effectiveness and computational cost depend critically on orchestration design. Across different tasks, role design, capacity assignment, and dependency construction jointly affect both solution quality and execution efficiency. Existing approaches automate parts of this design process, yet they often optimize these decisions partially or sequentially, and rely on execution-level feedback that provides limited credit assignment for local orchestration decisions. We propose LEMON (Learning Executable Multi-Agent Orchestration via Counterfactual Reinforcement Learning), an LLM-based orchestrator that learns to design efficient multi-agent orchestration. Given a task, LEMON designs a unified orchestration specification that composes customized agent duties, capacity levels, and dependency relations. To train the orchestrator, we augment the orchestration-level Group Relative Policy Optimization (GRPO) objective with a localized counterfactual credit signal that edits role, capacity, or dependency fields and applies the resulting reward contrast only to the edited spans. Experiments on six reasoning and coding benchmarks, including MMLU, GSM8K, AQuA, MultiArith, SVAMP, and HumanEval, show that LEMON achieves the best average performance among evaluated multi-agent orchestration methods while improving the accuracy-token trade-off.

A skilled live-commerce host is not merely a narrator, but a sales agent who converts viewer curiosity into purchase intent through expert product knowledge, emotionally intelligent response tactics, and entertainment that serves as a vehicle for product exposure. Yet no existing AI system replicates this. Conversational recommenders treat recommendation as a terminal act, while general-purpose LLMs hallucinate product claims and default to generic promotional templates that fail to engage or persuade. We present Ottilie, a sales-conversion-oriented virtual host for Chinese beauty live commerce that grounds product responses in a verified knowledge base, adapts discourse strategy to viewer intent through empathetic amplification, evidence-backed rebuttal, and humor-mediated deflection, and supervises all four output fields simultaneously within a structured generation framework. Experiments demonstrate gains of 14% on informativeness and 5% on factual correctness, with consistent advantages in tactfulness and viewer engagement. Ottilie has been deployed as an autonomous English-language host on Twitch, where the same fine-tuned model runs unattended on a single commodity GPU and promotes the company's own product; we report how the system adapted to a new language, market, and product without retraining, and what the deployment revealed about latency and adversarial chat.

Equipping LLM agents with reusable skills derived from past experience has become a popular and successful approach for tackling complex and long-horizon tasks. However, such lessons are often encoded as textual guidance that remains largely advisory, lacking explicit mechanisms for when and how to intervene in the agent loop. To bridge the gap, we introduce HASP(Harnessing LLM Agents with Skill Programs), a new framework that upgrades skills into executable Program Functions (PFs). Rather than offering passive advice, PFs act as executable guardrails that activate on failure-prone states and modify the next action or inject corrective context. HASP is highly modular: it can be applied at inference time for direct agent-loop intervention, during post-training to provide structured supervision, or for self-improvement by evolving validated, teacher-reviewed PFs. Empirically, HASP drives substantial gains compared to both training-free and training-based methods on web-search, math reasoning, and coding tasks. For example, on web-search reasoning, inference-time PFs alone improve the average performance by 25% compared to (multi-loop) ReAct Agent, while post-training and controlled evolution achieve a 30.4% gain over Search-R1. To provide deeper insights into HASP, our mechanism analysis reveals how PFs trigger and intervene, how skills are internalized, and the requirement for stable skill library evolution.

Harness evolution enables frozen language model agents to adapt to unfamiliar tasks by modifying the external programs that govern their behavior. For long-horizon tasks, each rollout is costly, and a limited interaction budget may yield few examples of effective behavior. Sparse, delayed feedback also makes it difficult for a coding agent to diagnose failures and determine which modifications will improve performance. We present DemoEvolve, which uses human demonstrations to guide harness evolution under limited interaction budgets. The coding agent examines demonstrations alongside its own rollout history, extracts strategies and the conditions under which they apply, and turns them into reusable harness components. We compare demonstration guidance with self-rollout evolution and augmentation with retrieved textual knowledge under the same evolution procedure and agent interaction budget. On held-out seeds, DemoEvolve improves mean capped final-round progress from 16.83 to 20.00 in Balatro and mean floor reached from 18.17 to 28.83 in Slay the Spire 2, compared with the respective base harnesses. These results suggest that demonstrations can help harness evolution make better use of limited interaction experience by providing concrete guidance for diagnosis and modification.

Knowledge graphs (KGs) have become the core backbone of numerous downstream tasks such as question answering and recommender systems. However, despite all this, KGs are often very incomplete. To perform zero-shot knowledge graph completion in unseen KGs, which have different relational vocabularies from those used for pre-training, KG foundation models (KGFMs) receive a wide range of attention. Existing KGFMs often perform training using random negative triples, which are constructed by replacing the head or tail entity of a positive triple with a random entity. However, these negative triples are often constructed with limited quality, providing weak supervision for KGFM training. In this paper, we propose a simple yet effective adaptive negative sampling approach, KMAS, to enhance existing KGFMs. KMAS constructs hard negative triples through the updated relation embeddings generated from the existing KGFM's relation encoder. To further adaptively align with the evolving capability of the KGFM during the training process, KMAS adjusts the ratio of hard negative triples dynamically throughout the whole training process: after a warmup phrase, it increases the ratio linearly and then decreases linearly. Extensive experiments are conducted over 44 data sets. Experimental results demonstrate that our proposed negative sampling method can enhance many SOTA KGFMs without requiring excessive additional time or memory consumption.

LLM-based agents are increasingly used to generate GPU kernels, but they often struggle to determine when an optimization is sound because its required code state and dependencies are implicit in expert implementations. We introduce KLineage, which learns this missing "when" knowledge from expert kernels: instead of relying on forward rollouts, KLineage walks expert implementations backward through validation-gated simplifications and reverses each accepted step into a reusable optimization skill. Each skill records not only the optimization intent, but also when to apply the optimization technique, including where it applies in code, what conditions made it valid, what effect it has, and what failures its assumptions avoid. A downstream LLM materializes these skills on new code surfaces under the same compile/correctness/profile gate. This guidance on when to apply each optimization can help downstream models to generate higher-performance kernels. On five expert workloads across two NVIDIA architectures, these lineage-derived skills serve as an effective optimization curriculum, exceeding recent memory-based LLM-kernel baselines in both final kernel quality and optimization efficiency under the same fixed budget. We also demonstrate that the KLineage framework extends beyond NVIDIA GPUs to Ascend NPUs. Our code is publicly available at https://github.com/ict-agent/klineage.

Reinforcement learning with verifiable rewards (RLVR) trains reasoning models without labeled trajectories, using groups of verifier-scored rollouts to explore alternative reasoning paths. Limited rollout diversity is a central bottleneck, typically addressed through adjustments to temperature, prefixes, or rollout selection. We identify the first token of the response as a structurally distinct target for diversification, largely overlooked in prior work. We find that the first-token distribution is sharply concentrated and only weakly related to downstream correctness, as lower-probability candidates can yield similarly accurate responses. Diversifying the first token can therefore broaden the reasoning paths explored within each rollout group with little loss in response quality. Motivated by this observation, we introduce REFT (Rollout Exploration with First-Token Diversification), a lightweight modification to RLVR. REFT samples first tokens uniformly from the policy's top-$N$ candidates and allocates rollouts evenly across the sampled tokens, leaving the rest of the pipeline unchanged. We evaluate REFT on eight models spanning multiple architectures and sizes (0.5B-14B), with mathematical reasoning and code-generation tasks under GRPO and DAPO. Across these settings, REFT consistently improves Pass@1, Pass@8, and Pass@64. It also outperforms competing diversification methods at every evaluated budget, incurring the lowest rollout cost.

As scientific literature grows rapidly and research increasingly involves AI agents, automated survey generation has become a key capability for both agents and human researchers. For such agents, a survey serves as a primary knowledge source of a field prior to research. Since pretrained models already encode broad knowledge of established work, these consumers benefit more from analytical depth than from breadth alone; moreover, unsupported claims, once ingested as knowledge, can propagate into downstream research. However, existing systems tend to overemphasize coverage and presentation; they suffer from limited analytical depth due to reliance on abstracts and isolated paper processing, and from unreliable citations due to imprecise retrieval and post-hoc grounding. We present DeepSurvey, an agentic generation system that addresses both limitations. To enhance depth, DeepSurvey extracts structured keynotes, models cross-paper relationships through clustering and comparative analysis, and integrates a code-agent subsystem to recover implementation-level details. To fortify reliability, it combines citation-graph expansion with hybrid filtering for topic-focused retrieval, enforces evidence-constrained analysis and writing, and deploys multi-granularity agentic refinement to validate citation--claim alignment. Experiments show that DeepSurvey achieves the highest content score (8.34/10) and citation quality (recall and precision gains of 25.3\% and 35.2\% over the strongest baseline), generalizes more robustly across domains, and is preferred by domain experts over human-written surveys (83.3\% in overall quality, 100\% in content depth). Moreover, when a coding agent uses a survey as its only literature source, the agent equipped with DeepSurvey achieves the best performance among human-written and baseline-generated surveys.

Natural-language skills let agents reuse task knowledge, yet deploying a long Markdown document makes the model interpret that knowledge anew on every call. We ask whether the behavior induced by a skill can be carried by a compact, trainable context. SoftSkill initializes virtual token embeddings from a skill document and optimizes a soft skill with next-token prediction while keeping the language model frozen. The resulting conditioning sequence can occupy the skill section or another supported prompt location. On Qwen3.5-4B, a 32-token soft skill improves over no skill by 7.6, 42.1, and 1.3 points on SearchQA, LiveMath, and DocVQA. It also exceeds the optimized textual skill on SearchQA and LiveMath by 4.5 and 12.5 points, respectively, while replacing skill documents of hundreds to thousands of tokens. The same approach improves multi-step agent execution: on Qwen3.6-35B-A3B, OfficeQA and ALFWorld rise by 8.2 and 14.1 points over no skill; on Qwen3.5-4B with aligned action decoding, ALFWorld success rises from 44/134 with the untrained initialization to 91/134 after training. These findings show that a skill document can serve as the starting point for a continuous control that improves both answers and actions without updating the backbone.

The Hitchhiker's Guide to Agentic AI is a comprehensive practitioner's reference for building autonomous AI systems, covering the full stack from first principles to production deployment. The central thesis: building great agentic systems requires understanding every layer of the pipeline, not just one. The book opens with the LLM substrate, covering transformer architecture, GPU systems, training and fine-tuning (SFT, LoRA, MoE), model compression, and inference optimization, as essential foundations. It then develops the alignment and reasoning layer: RLHF, PPO, DPO and its variants, GRPO, reward modeling, and RL for large reasoning models including chain-of-thought and test-time scaling. The second half is devoted to agentic AI proper: agentic training and trajectory-based RL, RAG and Agentic RAG, memory systems (in-context, external, episodic, and semantic), agent harness design, loop engineering, graph-based orchestration, and a taxonomy of agent design patterns covering security, red teaming, and gateway infrastructure. Inter-agent coordination is covered in depth: the Model Context Protocol (MCP), agent skills and tool use, the Agent-to-Agent (A2A) protocol, and multi-agent architectures spanning centralized, decentralized, and hierarchical topologies. The book concludes with agent development frameworks, agentic UI design, evaluation methodology (non-deterministic evaluation, reasoning collapse, LLM-as-Judge), production deployment, and the regulatory environment (EU AI Act, California SB 942) as an engineering requirement. Each chapter pairs theory with implementation guidance, executable notebooks, and references to the primary literature.

We identify a structured confound in Reinforcement Learning from Human Feedback (RLHF). Pairwise preference labels are intended to reflect the compared outputs, but they may also reflect the rater's state during annotation. Under sustained stressful or distressing conditions, raters' preferences may shift over time, so that preference data encode rater state alongside judgments about response quality. We argue that, if present, such shifts would differ from ordinary disagreement or random label noise. They would be state dependent, could be shared across annotators under similar conditions, and would not necessarily cancel during aggregation, reward modeling, and policy optimization. We propose rater state shift as a plausible and testable source of structured bias in RLHF preference data. This paper develops a hypothesis and an audit framework for studying this source of bias. We define rater state shift, rater state confound, and correlated rater state bias. We also propose survival level emotional authenticity as a candidate output signature, defined by lexical, pragmatic, discourse, and safety features whose reliability and validity remain to be demonstrated. We show that systematic rater state bias can survive aggregation and may enter the learned reward signal. We state five testable predictions, together with effect size thresholds for an initial audit, and note which require proprietary data. Finally, we present an audit protocol and pilot study plan that can be applied to publicly available instruction tuned models. We do not infer the training history of any specific deployed model.

Membership inference attacks (MIAs) test whether a candidate example was used to train a language model. Existing attacks on fine-tuned discrete diffusion language models (dLLMs) often aggregate reconstruction signals across many mask configurations, requiring repeated model evaluations. We propose Joint Uncertainty Guided Mask Probing (JUMP), an efficient MIA that exploits the ability of dLLMs to predict masked tokens in parallel. Using the pre-fine-tuning checkpoint as a reference, JUMP selects low-confidence positions, masks them jointly, and aggregates clipped target-reference reconstruction gaps. This focuses the attack on positions that reveal stronger membership signals from fine-tuning. After mask selection, all selected tokens are evaluated with one scoring query per model. Across six MIMIR domains, JUMP improves mean ROC-AUC over a prior multi-mask attack from 0.819 to 0.902 on LLaDA and from 0.851 to 0.942 on Dream. Including mask selection, it requires only three forward passes per example, compared with 32 for the baseline. We further extend JUMP to the target-only setting by replacing target-reference scoring with relative token preference, which compares the observed token with alternative predictions at the same masked position. Target-Only JUMP achieves mean ROC-AUCs of 0.609 and 0.638 on LLaDA and Dream.

Long-running autonomous agents plan, act, and judge their own completion without human intervention. When an agent grades its own work, self-evaluation bias takes hold: plausible changes are accepted as progress while real-world outcomes stagnate or regress. We name this failure mode the progress mirage and show, with controlled measurement, that it is a question of what the evaluator is grounded in. We built a testbed that holds the agent and its tool surface fixed and manipulates only the information-channel type of the evaluator that gates the loop. A world-state oracle, unfakeable in principle, is enforced by container and network isolation and verified at every run. Across 54 cycles a frontier agent claimed improvement every time, yet 56 percent had a measured delta of zero or below. Self-report was thus uninformative, and the self-verdict gate degenerated into accept-all, eroding the best deployed state it had reached by 19 percent. Even the strongest in-band judge, reading the full artifact text, the change diff, and its own verdict history, accepted cycles of which 44 percent were real-world regressions and rejected 38 percent of real improvements; the preregistered adversarial hypothesis that a strong judge closes the gap was rejected. On a boundary task whose success specification is verifiable from the artifact itself, the same judge's mirage vanished to zero and the gap collapsed within the registered threshold, showing that the gap depends on where the success signal resides. A sign-only variant returning only the acceptance verdict kept real-world output similar to full feedback (110.0 versus 113.0), locating the benefit in the gate's grounding rather than in feedback content. For open-ended objectives whose success signal lives outside the transcript, scaling up the judge is not enough; out-of-band evaluation with real-world access is a structural requirement.

Recent advancements in Computer Go, driven by AlphaZero and MuZero, rely heavily on Monte Carlo Tree Search (MCTS) to correct the errors of the neural network policy. While effective on massive computational clusters, this dependence creates a critical bottleneck on consumer-grade hardware, where the computational cost of tree management severely limits inference rates. Furthermore, without deep search, these models suffer from hallucination, proposing moves with high confidence that are strategically fatal. This paper introduces a novel Belief-Guided architecture that disentangles the Policy head from a distinct Belief head. Unlike traditional value functions, the Belief head acts as an internal simulator and independent critic, modeling epistemic uncertainty and strategic stability. By integrating memory mechanisms (Transformer/GRU) to handle long-term dependencies and the Ko rule, and utilizing a gating mechanism to filter overconfident policy errors, our model shifts the burden of intelligence from runtime search to parametric "intuition." Experimental results demonstrate that this approach significantly improves search-free win rates and reduces hallucination, enabling professional-level play on limited hardware where massive MCTS is infeasible.

Heterogeneous multi-agent LLM systems, where agents are powered by different model families, can outperform homogeneous configurations by reducing redundant reasoning patterns. Yet existing communication protocols either operate through text, discarding the sender's internal representations, or require architectural homogeneity for latent-level transfer. We identify the entity grounding problem in cross-architecture communication: cross-attention bridges that transfer continuous representations across different LLM families suffer from rare-token compression collapse, where entity identity is lost in the continuous bottleneck (bridge-only F1 ~30%). We propose XBRIDGE, a decode-free communication protocol that addresses this through two mechanisms. Lexical Anchor Mapping (LAM) maps the sender's original context tokens to the receiver's vocabulary, providing discrete entity anchors. A Latent Enrichment Bridge (LEB) lets the receiver query the sender's hidden states for contextual enrichment. The entity anchors ground the bridge's contextual signals to specific entities through the receiver's own self-attention. Across three model families (Llama, Qwen, and Mistral), seven benchmarks, and both communication directions, XBRIDGE outperforms text-based communication on all seven tasks for each model pair while achieving 11x lower latency, and in a same-architecture setting it also exceeds a KV-sharing baseline on six of seven tasks. LEB requires only 264M trainable parameters (3.8% of the receiver), is trained on a small balanced sample set, and adds negligible inference overhead.

As large language models serve ever more requests, cumulative inference cost is growing relative to the one-time cost of training. In typical serving, prompt prefill runs in parallel and is compute-bound, whereas autoregressive decode is sequential and memory-traffic-bound. Conventional width or depth scaling raises both costs together, since every added layer is evaluated in both phases and enlarges the weights read at each decode step. We instead ask whether additional learned computation can be allocated to continuation prediction while preserving prompt-wide primary computation and a single KV cache. We realize this with the Decode-Branch Transformer. Its primary path alone processes the prompt and writes the KV cache; the decode branch is omitted during prefill and activated only from the final prompt position onward, adding continuation computation without writing state or affecting the primary path. The paths share attention, MLP, and output matrices, using separate token embeddings with lightweight coupling. Grouped decode reuses loaded weight tiles and the primary KV cache across both paths, so the added arithmetic does not proportionally increase dominant memory traffic or decode latency. Across matched-token comparisons, Decode-Branch achieves lower validation loss across architectures and data settings. In MoE models, the primary and branch expert fan-outs become independent knobs for trading prompt cost, decode cost, and predictive quality. We study two expert-allocation regimes, holding prefill or decode computation fixed, and expose a prefill-decode-quality trade-off enabled by phase-specific expert allocation.

Spatial intelligence is becoming a foundation for embodied agents, robotic planning, and multimodal assistants. To improve the spatial reasoning ability of VLMs, existing work has mainly followed two lines. One line uses post-training methods, such as supervised fine-tuning and reinforcement learning. Another line adopts an agentic paradigm in which the model calls external spatial tools to gather intermediate spatial evidence. We study a complementary and underexplored route: Can a frozen VLM agent improve its spatial reasoning through \textbf{parameter-update-free self-evolution}, without depending on external expert spatial tools at inference time? We present \textbf{Spatial Memory Agent (SMA)}, an experience-grounded runtime memory framework that converts verified spatial experience into reusable transferable lessons. Specifically, SMA first queries the frozen VLM in a verifiable spatial environment, obtains a predicted answer and reward, and uses verifier-guided reflection to distill compact transferable lessons stored in memory cards. SMA further assigns each memory card a \textbf{Transfer Reliability Score (TRS)}, which is initialized uniformly and calibrated from later retrieval outcomes as visit evidence of future transfer reliability. During read-only deployment, SMA retrieves memory cards through semantic filtering and combined similarity--TRS ranking, allowing the retrieved memory to guide frozen model inference. Experiments across five representative spatial benchmarks show that SMA achieves the best macro-average accuracy for all four base VLMs and the best accuracy in most individual evaluations, establishing a practical parameter-update-free path for spatial self-evolution through reusable experience.

Pretrained large language models offer a practical foundation for learning useful behavior from few task-specific examples. We argue that current prompt and context optimization methods underuse the extensive knowledge and reasoning capabilities of trillion-parameter models. These capabilities can make adaptation more sample-efficient, more compute efficient and at no performance loss when organized around how human experts investigate failures. We formalize Policy Iteration with Human Feedback (PIHF), which makes this implicit procedure explicit for LLM agents to execute, and build its automated implementation, PIHF-MCP. Initialized from clinician feedback on rare-disease diagnosis, PIHF-MCP supplies the expert procedure, testing tools, review and persistent inquiry records to develop reusable task policies. Across general reasoning benchmarks (BIG-Bench Extra Hard, HoVer and LiveBench-Math), PIHF-MCP improved performance of the baseline model by 16.9, 22.2 and 4.7 percentage points, respectively. With a matched baseline model, development used about 1/5 of the labelled examples and 4% of the task rollouts reported by a previous SOTA in-context optimizer, making it about 9 times faster and 3 times cheaper at comparable or higher scores. In a low-data rare-disease diagnosis setting, policies developed from previous SOTA prompt optimizers trailed a previously published PIHF-developed system on every held-out cohort (on average 16 percentage points). These findings support a route to more efficient inference-time scaling: PIHF-MCP develops reusable policies from a few examples that improve performance on unseen cases and across models. Because each policy comes from an explicit, recorded investigation, the process also keeps humans in the loop and enables ownership and learning, making it well suited to high-stakes decisions.

Multivariate time series forecasting (MTSF) is a fundamental task in many real world applications. Existing patch based forecasting methods generally fall into three categories: fixed partitioning, multi-scale partitioning, and extendable partitioning. Fixed partitioning often breaks meaningful temporal boundaries, multi-scale partitioning may introduce redundant representations across scales, and extendable partitioning improves flexibility but still lacks an explicit mechanism for organizing semantic structure and modeling interactions among heterogeneous temporal patterns. To address these limitations, we propose SCPaT, a Transformer based framework built on semantic structured partitioning. SCPaT first decomposes input sequences into semantically consistent units through adaptive semantic unit generation, then constructs a dynamic semantic graph to model directed dependencies among these units and organize them into higher order semantic blocks. Based on these structured representations, an importance aware routing mechanism adaptively dispatches different semantic blocks to different experts for customized modeling. Extensive experiments on 12 real world datasets demonstrate the effectiveness of SCPaT.

Robust POMDPs (RPOMDPs) generalize classical POMDPs to the setting where exact transition probabilities are not known -- rather, they are only known to belong to some uncertainty set of values. In this work, we study the problem of solving RPOMDPs with general omega-regular objectives, which subsume a broad class of objectives such as reachability, safety, and linear temporal logic (LTL) objectives. We show that, for (s,a)-rectangular RPOMDPs with polytopic uncertainty sets, the problem of solving RPOMDPs under omega-regular objectives can be reduced to solving partially observable stochastic games (POSGs) under omega-regular objectives. Moreover, we show for the first time that reductions can be constructed in both directions, establishing the semantic equivalence between (s,a)-rectangular RPOMDPs with polytopic uncertainty sets and POSGs. This allows us to derive a range of new computational complexity results, including both upper and lower complexity bounds, on solving RPOMDPs with different omega-regular objectives. As a corollary, we also derive new computational complexity results for RMDPs.

Bias in human-agent interaction can manifest not only through hostile language but also as benevolent bias, whereby unequal treatment hides behind a warm, positive tone. To make it detectable, we operationalise benevolent bias along two dimensions, tone and treatment, yielding three classes: neutral support, overt bias, and benevolent bias. Building on these definitions, we construct BENEVDIAL, a class-balanced corpus of 362,880 multi-turn support dialogues spanning user and agent demographics, roles, and generators, to support controlled evaluation. We then test two detector families on it: off-the-shelf safety detectors and prompted large language model (LLM) judges. Our findings reveal a notable detection gap: off-the-shelf detectors reliably flag overt bias yet largely fail to identify benevolent bias. LLM judges improve sensitivity when guided by explicit detection criteria, but this comes at the cost of increased misclassification of neutral supportive statements as benevolent bias, a tendency that is further exacerbated by the presence of demographic context. These findings suggest that fair monitoring of human-agent dialogue must look beyond surface cues to whether the agent's treatment is disparate.

Multi-agent systems enable complex reasoning and tool use by coordinating agents that divide roles and refine candidate solutions. Existing methods typically update individual agent responses or treat a complete trajectory as one training example. However, these methods may produce misleading policy updates because they assign the same final outcome to responses or trajectory segments that may play different roles in different team decisions. This is because treating each response as an independent update may separate outputs that jointly determine the next action, while treating an entire trajectory as one update may combine decisions made after different observations. These limitations call for a policy update defined at the level of a team decision, outputs that lead to the same state transition are optimized under a shared objective. In this paper, we propose Setwise Relative Policy Optimization (SRPO) for multi-agent systems. We represent the outputs used together to produce one state transition as an active set. A singleton set covers division of labor, while a larger set covers joint co-evolution, and the set composition can change across decisions. SRPO assigns a shared advantage to each active set, clips the combined policy change, and normalizes its scale according to the set size. This ties each update to the decision that produced the next state. Experiments on mathematical reasoning and multi-turn search demonstrate the effectiveness of SRPO across both tasks. Additional ablation studies analyze the normalization choice and training behavior under changing active-set sizes.

Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language feedback have emerged as a leading paradigm. In this paper, we identify limitations in two stages of existing textual gradient approaches: gradient extraction and gradient aggregation. In gradient extraction, previous works select a target prompt without verifying whether modifying it resolves the failure, and derive gradients without agent-level supervision over the corresponding agent's intermediate output. In gradient aggregation, individual gradients are randomly grouped and concatenated, often mixing unrelated failure modes and producing prompts that fail to generalize. To address these limitations, we propose AgentGrad, a prompt optimization framework for multi-agent systems based on sequential intervention and semantic textual gradient abstraction. For each failure, sequential intervention modifies the behavior of one agent at a time to identify the target agent whose modification resolves the failure. The modified output of the target agent then serves as agent-level supervision for extracting a fine-grained gradient. Semantic textual gradient abstraction clusters semantically similar gradients to prevent mixing unrelated failure modes, and abstracts each cluster into a generalized gradient that captures the shared corrective pattern. Experimental results show that AgentGrad achieves state-of-the-art performance across five MAS benchmarks while reducing wall-clock optimization time by $2.5\times$ and optimization cost by 21.8\% on average compared to the next-best baselines.

The increasing deployment of AI agents in long-horizon tasks yields massive execution logs. Diagnosing failures within these records is crucial for reliability, as it transforms outcome-level signals into actionable interventions. The sheer scale of the data renders human review impractical, driving the need for automated root-cause attribution (RCA). However, automated RCA methods using LLMs suffer from low diagnostic accuracy, especially as execution traces grow larger. They struggle because relevant information is often sparse, distributed across distant actions, and disconnected from the visible failure, reducing root-cause attribution to a massive search problem. Existing RCA methods typically rely on one-shot LLM judgments to diagnose failures from execution traces. While effective for shorter trajectories, these judges tend to settle on a plausible diagnosis early, leaving critical evidence in longer traces unexamined. We introduce Continual Search, an iterative framework that nudges the judge, over successive turns, to keep searching for unresolved diagnostic evidence. We evaluate Continual Search across four existing RCA benchmarks. Recognizing the lack of massive execution traces in current benchmarks, we introduce MegaRCA-Mix to evaluate RCA at scale. MegaRCA-Mix provides a challenging testbed of 50 human-annotated failure trials spanning long-horizon, execution-heavy tasks. Across multiple benchmark suites and model families, Continual Search consistently improves attribution performance. On MegaRCA-Mix, for example, it improves Opus-4.8's F1 score by 29\%, from $0.471$ to $0.608$. More interestingly, within the same model family, lower-tier models can even surpass their higher-tier counterparts, demonstrating that effective search supersedes raw model scale.

Continuous reasoning has emerged as a promising way to improve reasoning in large language models (LLMs). Yet we still lack a clear principle for deciding what a latent state should preserve. Reasoning by superposition shows that a single latent state can encode several search alternatives and expand them in parallel. We ask how those states should be weighted as reasoning proceeds. A natural choice is to preserve only the states active at the frontier step, since keeping every reached state appears to spread a limited hidden width too thin. We show that the opposite can hold. When later computation draws on several reached states, a cumulative state can guide attention correctly at a smaller hidden width than a frontier state that stores fewer states. At the same width, the cumulative state therefore keeps more intermediate states available for later reasoning. More generally, equal cumulative weights are optimal when future queries are unknown and remain close to the best task-specific weights when those queries are known. Experiments with two-layer and GPT-2 Transformers reproduce the predicted width advantage and show that unequal weights fail first on the states that receive the least weight. This suggests a important principle: keep reached states equally weighted, and restore equal weights as computation proceeds.

Systematic generalization, the ability to solve novel problems by recombining known atomic elements, is central to human intelligence but difficult to study rigorously under controlled settings. Existing studies therefore rely on simplifications such as elemental composition, productivity-based tests, and action-explicit goals, which make systematic generalization easier to study but omit some essential aspects of this capability. To characterize what these simplifications miss, we adopt a reasoning-centered lens and introduce TranSGrid, a testbed that brings deductive, inductive, and abductive reasoning together within a unified task. Experiments with seven Transformer models on 4,800 TranSGrid instances show that all models perform much worse on TranSGrid than on a held-out test set: the largest model solves 79.6% of the test set, but only 55.3% of TranSGrid and 15.8% of the hardest subset. The gap remains within the training length range, showing that productivity alone is not sufficient to evaluate systematic generalization. Additionally, we reintroduce the other two simplifications into TranSGrid: one variant limits interactions among action effects to approximate elemental composition (reducing the inductive demand); the other makes goals action-explicit (reducing the abductive one). In both, solve rates return to roughly the test set level, showing that either simplification alone is enough to reduce TranSGrid to an ordinary held-out test set. Together, our results show that existing tasks reduce either or both of the inductive and abductive demands of systematic generalization, and that comprehensively measuring this capability requires a task that involves all three forms of reasoning.

Accurate flood forecasts several days in advance are essential for flood control, water-resource management, and emergency response. A central challenge is to produce high-resolution forecasts across continental domains where hydrological behavior varies widely from place to place. We developed Hapi, a U-Net Swin Transformer that uses fine three-dimensional patches and hierarchical shifted-window attention to forecast river discharge, surface runoff, snow water equivalent, and soil wetness index across the contiguous United States. The model produces medium-range forecasts (24--72~h) at $0.05^{\circ}$ resolution with adaptive task weighting and required only 0.11 seconds for a four-variable 72-h CONUS forecast on one A100 GPU. In a held-out 2024 potential-skill evaluation with ERA5-Land inputs prescribed over the forecast horizon, Hapi achieved the highest F1-score for floods in 20 of 21 comparisons across seven GloFAS return periods and three forecast leads. Independent validation against observed daily discharge at 3{,}881 U.S. Geological Survey gauges showed that Hapi achieved the highest median Nash--Sutcliffe efficiency at every lead, supported by regional-cluster bootstrap intervals. In a matched 24-h comparison of loss formulations, adaptive task balancing produced the lowest discharge errors and the highest F1-score for floods.

LLM-as-a-judge scales evaluation, but reasoning judges are slow and costly. We study JEV-as-a-Judge: evaluation with JEV, a decision-only judge that returns label probabilities instead of text, and whose confidence decides whether to accept its verdict or escalate to a reasoning judge. Against sixteen generative and reward-model judges, with blinded human adjudication, JEV comes within three points of GPT-6 wherever a verdict can be read off the text, at 0.36% of its fee and a 0.15-second median latency, and falls behind where the verdict must be derived, as in math, code, and logic. Its confidence marks this boundary. With a threshold frozen in advance, accepting confident verdicts and escalating the rest is 0.9 points more accurate than GPT-6 on 1,610 held-out pairs at 41% of its fee, and in a pre-specified live test on two new workloads the cascade matches GPT-6's accuracy exactly. Confidence routing weakens on style-adversarial pairs and reference-free prose; we close with a simple recipe for validating thresholds locally.

Pretrained ASR systems perform poorly on noisy Broadcast Police Communication (BPC), hindering efforts to understand police decision-making. Pseudo-labeling offers an unsupervised path to improve ASR without expensive human labels, but the efficacy of this approach on very noisy domains is not known. In this work, we systematically assess the opportunities and limits of pseudo-labeling to adapt foundation ASR models (Whisper and Qwen3-ASR) to noisy BPC domain corpora from Baltimore and Chicago. We demonstrate that existing internal confidence metrics (log-probabilities and STAR scores) fail to distinguish between high and low quality BPC pseudo-labels, and we introduce an external LLM-as-a-judge filtering paradigm that leverages parametric knowledge to discard contextually implausible transcripts. Our LLM-judging filters more aggressively than internal metrics and significantly reduces WER of the pseudo-labeled training sets across the Baltimore and Chicago BPC corpora, though a substantial gap remains relative to an oracle filter. We also introduce a new cross-model pseudo-labeling paradigm where one model is finetuned with pseudo-labels from the other, and we identify this method as a promising direction for future pseudo-labeling work.

Audio large language models allow users to interact with the model through speech. When an input recording is too degraded, the model may misinterpret the user's query and respond based on an incorrect transcription. In this paper, we study model-conditional transcription reliability: whether an Audio LLM can recognize when its own transcription is unreliable. We first prompt the Audio LLM to assess whether its own transcription would be reliable, and find that the model is a poor judge of its own transcription reliability: in most cases, it predicts that its transcription will be reliable. We find that existing approaches, including speech quality predictors, audio LLM generation uncertainty, and transcript-conditioned WER estimation, provide limited signals for detecting transcription failures. In contrast, we discover that transcription reliability is strongly represented in the model's audio-encoder representations. Based on this observation, we devise a lightweight reliability predictor that operates on representations extracted by the frozen audio encoder and predicts the reliability class before generation. The reliability predictor can trigger a clarification request from the user when their voice query is predicted to be unreliable, while allowing reliable queries to proceed without modifying the underlying Audio LLM. Our predictor achieves 81.10% in-domain and 78.09% cross-domain macro-F1 scores, outperforming the strongest baselines by 10.33 and 11.93 points, respectively. Finally, we show that reliability labels can transfer across Audio LLM families, and that transfer performance is closely related to the alignment of their model-specific reliability boundaries.

Large Language Model (LLM) pollution occurs when synthetic responses contaminate data intended to capture human behavior. High deployment costs have so far limited the risk posed by autonomous survey agents. However, open-weight models paired with open-source agentic frameworks may have removed this barrier. We compared the performance and detectability of nine agent configurations, ranging from fully open variants to closed commercial ones. Each agent autonomously completed a survey containing multiple response types yielding various detection checks. Fully open agents ran locally without usage fees and performed competitively with commercial alternatives. Open and commercial agents failed different sets of checks, and no single check reliably detected all agents, but open-text responses discriminated best between agents and humans. These findings identify fully open agents as a distinct risk for LLM pollution and support multilayered detection strategies emphasizing open-text analysis.

Validating an LLM-as-a-judge requires estimating its agreement with humans, yet annotation budgets rarely allow every item to be multiply labeled. We prove that this overlap sparsity is the first-order determinant of wrong deployment decisions: at 5% pairwise overlap, wrong-decision rates reach 25% and the probability of selecting the wrong best judge among ten candidates is 65%. The two actionable levers are overlap quantity and allocation. For quantity, we derive a minimum-overlap formula showing $\rho \geq 0.25$ suffices for non-borderline judges while borderline cases remain fundamentally hard. For allocation, a zero-cost stratified scheme halves false-rejection rates relative to random sampling when strata are informative. We validate on 10 LLM judges across four evaluation matrices spanning visual assessment, causal reasoning, and summarization.

Multi-agent workflows require task-relevant information to be shared across agents, while irrelevant, stale, unverified, or incompatible information must remain isolated. We call this task-conditioned scope of information a collaborative memory boundary. Workflow topology determines which intermediate artifacts are applicable to which downstream workers and when they cease to be valid, thereby providing a structural stress dimension for sharing and isolation. Existing memory benchmarks primarily evaluate retention and retrieval, whereas multi-agent benchmarks emphasize coordination and end-to-end completion, leaving topology-conditioned memory boundaries largely unmeasured. We introduce CoMemBench, an execution-grounded benchmark for collaborative memory sharing and isolation across multi-agent workflow topologies. It constructs 800 composite workflows across four domains from source-grounded dependency graphs, with node-local specifications, verifiable artifact handoffs, native evaluators, and matched isolation challenges. CoMemBench measures workflow completion, verified node progress, required-handoff reliability, isolation robustness, and token cost. Experiments reveal a sharing-isolation trade-off: broader context improves information availability but can weaken isolation, while system rankings shift across topologies and artifact violations.

LLM-based multi-agent systems (MASs) have shown strong potential for solving complex tasks, but most assume that task requirements are sufficiently specified before execution. In practice, user requests are often incomplete, and additional requirements may only become clear during reasoning, tool use, or execution. We refer to such problems as progressively specified tasks. To systematically study this setting, we introduce ProgSpec, a benchmark that evaluates final outputs against requirements explicitly stated in the initial request and additional requirements supported by the available task evidence. We further propose RepoMAS, an issue-driven multi-agent framework inspired by open-source project management. RepoMAS records newly discovered requirements, conflicts, and failures as structured Issues and uses them to revise the task specification and execution structure during problem solving. Across ProgSpec and five existing benchmarks, RepoMAS achieves the best performance. Further analyses show that its issue-driven revision and repository maintenance mechanisms consistently contribute to performance. These results highlight the importance of allowing MASs to revise not only how a task is solved, but also revise their explicit representation of task requirements during execution.

Language models (LMs) often hallucinate by committing to confident answers rather than abstaining, even when they do not have enough information to answer reliably. A large body of existing work mitigates hallucination through detection or abstention mechanisms, but leaves open how models internally arrive at the decision to commit or abstain in the first place. We study this decision through mechanistic analysis, framing hallucination as unsupported commitment: the model commits despite exhibiting signals of unanswerability. Using causal gating, we identify a Commit-Abstain Circuit (CAC), a sparse, causally localised subset of attention heads and MLP sublayers underlying this decision. Across ten LMs (3B-14B) from five families and three benchmarks, the CAC exhibits a recurring accumulate-yet-undercorrect pattern: commitment-promoting components build up commitment in earlier layers, while abstention-promoting components act later as corrective signals that are often insufficient to overturn the accumulated commitment. Building on this finding, a lightweight policy trained on CAC activations improves decision accuracy by 12.2 points over the model's intrinsic commit-abstain margin, reduces false abstentions by 2.5 times, transfers to unseen benchmarks, and extends to larger models (27B-35B). The CAC is both diagnostic, clarifying how models overcommit, and practical, enabling improved abstention decisions.

Multiple agents may often conflict in an organization: for example, one coding agent changes an interface in a repository, but another continues to develop on the old version where existing tests become stale. A conversation can resolve the episode, but when the participants change, what makes the lesson continue to govern the team? We introduce Relic, which turns recurring collaboration failures into organization-owned, executable protocols. Members reflect on visible work, propose rules, and govern their adoption. Adopted protocols bind triggers, responsibilities, required evidence, and execution consequences to the runtime, while remaining open to revision and retirement. In one traced case, repeated integration friction produces an interface-review rule that governs later pull requests and is revised as work continues. Across 360 controlled runs over ten software workloads and three models, Relic raises complete-contract delivery from 14.06% to 19.76% (+5.71 percentage points) over a matched structured team without the protocol lifecycle, improving all four verified production endpoints in every model stratum. Under fresh-member transfer, behavioral correctness is 25.4% with no inherited protocol, 34.6% with the same rules provided as readable text, and 41.2% with executable bindings, a +6.5-point advantage over text alone. On the full CooperBench benchmark, after excluding 183 broken benchmark pairs, Relic achieves 371/469 (79.1%), establishing the best reported result among peer-structured systems. On the 47-pair same-model subset, Relic also exceeds Solo (28/47 vs. 26/47), reversing the coordination loss exhibited by the official peer baseline. Together, these results show how collaboration experience can become persistent organizational state that remains useful beyond the members who created it.

Self-evolving agent harnesses continually update persistent components such as memory, prompts, skills, and tools. We call this process harness evolution. However, such evolution could introduce unexpected safety risks. Existing work studies harness misevolution and validates candidate harnesses or attributed individual component updates, leaving safety analysis of cross-component update interactions largely unexamined. To address this gap, we study compositional safety failures in harness evolution, where interactions among individually safe and utility-preserving component updates can produce undesirable or unsafe agent behavior, revealing a safety risk intrinsic to harness evolution. Across three safety-related benchmarks, we identify 43 pairwise and 18 irreducible 3-way compositional safety failures. Conventional solution incurs combinatorial complexity in validating cross-component interactions, leaving the safety checking impractical as the harness evolves. To solve this, we introduced a typed hypergraph that represents component states as nodes and safety-relevant higher-order interactions as hyperedges. When the harness changes, the hypergraph updates only the interaction neighborhood of the changed states rather than reconstructing the global composition space. Building on that, we develop a hypergraph-guided runtime monitoring mechanism. Experiments show that our method effectively mitigates compositional safety risks while preserving task utility and reducing interaction-checking costs, and further reveal an empirical safety-utility-cost trade-off across different safety mechanisms.

On-policy distillation (OPD) trains a student model on its own trajectories using dense token-level feedback from a stronger teacher model. Since each update is conditioned on the reasoning prefix already generated by the student, the prefix also shapes how effectively teacher feedback is converted into learning. We find that shared prefixes can lead to weak token-level updates, a phenomenon we call Prefix-Induced Supervision Attenuation (PISA). This attenuation arises in two common cases. (i) High student confidence can weaken corrective gradients even when the teacher disagrees. (ii) Tokens that rely on earlier reasoning can receive learning signals as weak as those for simple local continuations. To solve this problem, we propose Trajectory Dropout, a simple training-time intervention that exposes these weakened signals. The student first performs a standard full-context rollout to generate a complete trajectory. During training, we randomly drop a certain proportion of the student's reasoning trajectory, while the teacher continues to observe the complete trajectory for token-level supervision. This intervention strengthens corrections for overconfident predictions and introduces additional supervision at prefix-sensitive positions. Trajectory Dropout consistently improves average performance across teacher--student model pairs of different scales and six mathematical reasoning benchmarks, while also yielding gains on two out-of-domain benchmarks. It can also be flexibly integrated into existing OPD variants with negligible computational overhead, further improving their performance. These results demonstrate that Trajectory Dropout provides a simple mechanism for strengthening token-level supervision across model scales and OPD objectives.

Executable environments are critical for post-training agents on tasks that require tool use and multi-step interaction, but constructing executable tasks together with their environments remains difficult to scale. Skills provide reusable domain knowledge, operational procedures, and tool-use instructions, but a substantial gap remains between the information contained in a skill and a concrete, challenging task with a complete executable environment. To address this gap, we introduce Skill2Env, a capability-oriented framework that starts from a skill and uses agent capability demands to guide task and environment synthesis. Skill2Env represents these demands through reusable difficulty patterns and instantiates them into task blueprints that specify objectives, challenges, environment facts, information boundaries, and acceptance criteria. These blueprints guide the joint construction of task instructions, execution substrates, workspaces, and rubric-based evaluators around source skills. We further propose Iterative Task Hardening, which uses solver execution evidence to identify insufficiently challenging task designs, strengthen or extend their difficulty-pattern instantiations, and revise the corresponding blueprints and environments. Using 1.5K high-scoring trajectories generated from Skill2Env environments for supervised fine-tuning, we observe consistent improvements across a broad range of agent benchmarks, demonstrating the effectiveness of capability-oriented environment synthesis for agent post-training.

Understanding where and why Graphical User Interface (GUI) agents fail is essential for building more reliable systems, yet current evaluation relies on step accuracy, a metric that treats each screen independently and overlooks the underlying structure of GUI environments. This leads to two critical blind spots: (1) functionally equivalent screens are evaluated in isolation, obscuring systematic failure patterns across shared screens; and (2) the long-tailed GUI distribution renders failures on rare but critical screens invisible under standard metrics. To address these issues, we propose \textbf{GUITAR}, a state-centric diagnostic framework that performs structured failure analysis over both states and transitions, using a State Transition Graph (STG) by mapping visually diverse screens to shared functional states. Across 8 agents and 6 tasks from AndroidControl and Mind2Web, GUITAR reveals that 60.4\% of failures occur in 20\% of states, localizing errors to a small set of bottlenecks. Bottleneck-targeted guidance improves SR by 2.8\% and retains a 1.88\% average gain across 7 agents under three-fold trajectory-held-out evaluation with fully automatic STGs. These findings demonstrate the diagnostic and actionable value of structure-aware evaluation within the evaluated mobile and web tasks. Code is available at https://github.com/sqzhang-lazy/GUITAR

The use and applicability of artificial intelligence (AI) in medical research and clinical practice has received increasing attention in the literature over recent years. The emergence of large language models (LLMs) has expanded discussions in regards to applications of AI within healthcare. While traditional deep learning based AI applications in medicine have often focused on specific and defined tasks, LLMs offer broader capabilities and flexibility in working with available data,. At the same time of writing, the integration of LLMs into medical settings raises important questions regarding their reliability, accuracy, transparency, safety, and appropriate role in a medical setting. This text presents and discusses recent talks and articles concerning the application of LLMs in medicine, with particular emphasis on their potential utility in research and clinical practice. It considers both the opportunities offered by these technologies and the challenges associated with their implementation, aiming to provide a perspective on the current and emerging role of LLMs within the medical field.

Language-conditioned trajectory generation is here, but its evaluation has not kept pace. Existing pedestrian trajectory metrics compare trajectories with real-world human data. This does not scale to text-to-trajectory generation across diverse contexts, as collecting human trajectories for every scenario is costly and infeasible. Moreover, pedestrian behavior is heterogeneous and context-dependent, with no single metric as the correct answer, and current evaluation frameworks are not transferable to this domain. These challenges make scalable, reliable evaluation difficult. We introduce STRIDE, the first framework for evaluating context alignment between scenario descriptions and pedestrian trajectories. STRIDE addresses these challenges through three design choices. First, we derive our VRDST evaluation protocol from sociological theories to define a complete evaluation space. Second, it decomposes high-level context into scenario-adaptive behavioral questions. Third, every question is resolved against a deterministic measurement tool library that yields reproducible answers. Together, STRIDE enables complete, verifiable, automated, and scalable evaluation across diverse contexts without requiring human trajectory data. We instantiate STRIDE in the crowd domain as STRIDE-Bench, comprising 1K scenarios, 6K behavioral questions, and 11K measurements with calibrated expected answers across 30 real-world maps. Comprehensive human validations show that STRIDE-Bench is consistent with human behavior and judgment, achieving 80% human agreement. We further evaluate several text-to-trajectory models, finding limited context-alignment capability and persistent challenges in fine-grained context conditioning. We believe that the STRIDE framework provides a first step toward principled evaluation of context-aligned pedestrian trajectory generation.

In 2023, AI models answered from training data and hallucinated when it ran out, and businesses were told to seed that knowledge. Models' training knowledge has since given way to live web search, and the advice followed it there: answer-engine optimization, or AEO, now tells businesses to scatter breadcrumbs across forum threads, listicles, and off-site citations, so AI engines are likelier to surface and recommend them. But being surfaced is no longer enough: an agent opens the results and reads them before deciding, and one buyer question sends it through several rounds of search and fetch. What decides the outcome at this drill-down step is whether the agent can fetch and read the business's own site: agent experience (AX). We argue that AX is the new AEO. We run 37,927 agent journeys, each a buyer question about a business, across four independent harnesses over 1,056 real businesses, matched on fame, prior model knowledge, and two AEO proxies, then split based on their AX level. Only 7-10% of the finished answer comes from the model's training knowledge, whether or not the site is readable. Agent-ready businesses have answers built from their own pages 78% of the time against 56% and are clearly recommended 1.9x more often, while every grounded answer about a not-agent-ready business costs the agent 64% more. Holding business, harness, and question fixed, answers built from the site are 41% more accurate. The dominant failure is not fabrication but omission: web-built answers are 3.7x more likely to contain none of the facts the buyer asked for. Baselines differ sharply across the four harnesses, with clear-recommendation rates varying sevenfold from stack to stack, yet the effect holds in every one. In the agentic web era, being readable beats being talked about, and improving a site's AX is the strongest lever a business has.

Large-scale routing problems are difficult to solve efficiently as their search spaces grow rapidly with problem size. Existing approaches primarily improve the optimization procedure itself, often at increasing computational cost. We instead shift the focus to a useful initialization that can be refined into a high-quality solution with limited downstream refinement. We propose Just Initialize, a training-free and solver-agnostic initialization component for large-scale routing optimization. Just Initialize compresses a large routing instance into a compact surrogate space, optimizes its global routing structure, and recovers the resulting solution as an optimization-friendly starting point in the original space. Extensive experiments on Traveling Salesman Problems (TSPs), Capacitated Vehicle Routing Problems (CVRPs), Vehicle Routing Problems with Time Windows (VRPTWs), and Prize-Collecting Traveling Salesman Problems (PCTSPs) demonstrate that Just Initialize achieves high-quality solutions comparable to or better than state-of-the-art methods while substantially reducing computational cost across instances ranging from 1K to 100K nodes, including an average speedup of approximately 70$\times$, sub-second runtimes on 10K-node instances, and runtimes within tens of seconds on 100K-node instances.

Rare-disease diagnosis is a long-tail reasoning problem: phenotypes are incomplete, individual disorders are sparsely documented, and relevant evidence is distributed across ontologies, gene annotations, and biomedical text. Language models consequently favor common conditions, miss rare candidates, or produce plausible but invalid names. We introduce RareDx, which couples controlled evidence use with knowledge-graph-grounded policy optimization. RareDx-Harness normalizes heterogeneous records into one ranked-diagnosis task and compares direct inference, static retrieval, adaptive tools, and structured phenotype-gene-disease reasoning over a shared knowledge layer. The training pipeline combines Top-10 post-training with RareDx-KGPO, our knowledge-graph-grounded policy optimization method. Its reward projects predictions into a canonical disease graph and integrates curated graded relevance, ontology proximity, biomedical similarity, and phenotype consistency. Vocabulary and output-budget constraints prevent dense partial credit from rewarding fabricated or overlong differentials. Across eight benchmarks, the complete RareDx system centered on Qwen3.5-9B reaches 38.34 macro Hit@10, 1.60 points above GPT-5.5 under the archived protocol; a disjoint validation-selection audit retains a 6.80-point routing gain over Direct on held-out cases. The 27B system reaches 23.53/36.56/40.76 at Hit@1/5/10. Controlled ablations show that retrieval is not uniformly helpful and that controlled routing is central to the gain. These results indicate that structured medical knowledge can turn a compact model into a competitive diagnostic ranker across heterogeneous long-tail settings in clinical practice.

When recalling lists of concepts (e.g., animals) during the semantic fluency task (SFT), both humans and large language models (LLMs) organise their output into clusters of related items (e.g., sea animals) that are punctuated by strategic switches between clusters. In humans, this pattern can be explained by a semantic foraging process, whereby distinct neural and behavioural signatures accompany within-cluster production ("exploit") and between-cluster switching ("explore"). Whether LLMs likewise represent these two search regimes within their internal states is unknown. Here, we apply a range of mechanistic interpretability techniques to provide evidence for this. In Study 1, we use the Jacobian lens (J-lens), which maps intermediate-layer residual-stream representations to token-level activations, to show that concept-level activations predict switching. First, we find that switching coincides with low next-token activations. Moreover, the probability of switching rises as the set of strongest J-lens activations (the J-space) becomes depleted of items from the category currently being produced, analogous to explore-exploit decision-making during patch foraging. We then show that middle-layer J-lens activations of abstract category-related labels (e.g., "water") increase in anticipation of switching into that category. We confirm these representations to causally influence switching by deriving steering vectors that target category switching. In Study 2, we identify generic residual stream directions that are activated during and in anticipation of switching. By steering activations along these directions, we bias increased or decreased rates of switching. Our study extends the semantic foraging framework to artificial intelligences and provides evidence that LLMs maintain distinct representational signatures for exploration and exploitation as they verbalise conceptual information.

Continuous diffusion generates complete reasoning solutions through iterative refinement in latent space. We introduce the Continuous Embedding Diffusion Reasoner (CEDR), an ELF-based training and inference recipe. Our experiments show that accurate decoding alone does not ensure strong reasoning performance. We therefore learn compact representations from multiple layers of a strong autoregressive teacher. Their decomposition also enables asynchronous denoising at different rates. We show that prompt encodings need only preserve the information required for the correct text-conditional score, rather than exactly match teacher features, and use a staged curriculum to learn a compact prompt encoder that replaces the teacher Transformer at inference. We adapt DiffusionNFT to learned self-conditioning guidance and incorporate gold-solution endpoints to supplement sparse rewards. Our supervised models outperform reported results from recent continuous-diffusion baselines at comparable backbone scales on mathematical reasoning and HumanEval code generation. With a 638M-parameter denoising backbone and learned prompt conditioning, post-NFT CEDR-L achieves 63.74% pass@1 on GSM8K and 24.6% on MATH500 at 64 denoising steps, and 32.85% on HumanEval and 30.18% on HumanEval+ at 128 denoising steps. Code will be available at: https://github.com/chengxiang/CEDR.

Risk-sensitive reinforcement learning (RL) is crucial for maintaining reliable performance in high-stakes applications. While traditional RL methods aim to learn a point estimate of the random cumulative cost, distributional RL seeks to estimate the entire distribution of it, leading to a unified framework for handling different risk measures. However, developing policy gradient methods for risk-sensitive distributional RL is inherently more complex as it often involves finding the gradient of a probability measure. This paper introduces a new distributional policy gradient framework for risk-sensitive RL, where we derive an analytical gradient of the probability measure of the cumulative cost. For practical implementation, we further design a categorical distributional policy gradient algorithm (CDPG) that approximates arbitrary distributions using a categorical family supported on fixed points. Using Conditional Value-at-Risk (CVaR) as the objective, we prove that the proposed CDPG converges to stationary points and establish its iteration complexities under inexact policy evaluation. Through experiments in a stochastic Cliffwalk environment, we demonstrate the effectiveness of the proposed algorithm and highlight the benefits of incorporating risk sensitivity into distributional RL.

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by retrieving relevant information from external knowledge bases to provide more accurate, contextually informed, and up-to-date responses. However, this reliance on external knowledge introduces significant security vulnerabilities, as many RAG systems (e.g., Google Search) rely on large and unsanitized data repositories (e.g., Reddit). In this paper, we unveil a novel threat in which attackers steer the RAG system's response by injecting malicious passages into its knowledge base. When a user's query contains attacker-specified trigger words, the RAG retrieves and refers to these malicious passages, enabling the attacker to steer the response without altering the user input or modifying the RAG weights. BadRAG operates in two phases: (i) malicious passages are optimized to be retrieved exclusively when trigger words appear in user queries; (ii) these passages are meticulously crafted to achieve adversarial generation objectives, including denial of service, sentiment manipulation, context leakage, and tool misuse. Our experiments show that injecting just 10 malicious passages (0.04\% of the external corpora) achieves a 98.2\% retrieval success rate and increases negative response rates from 0.22\% to 72\% for queries containing triggers.

Knowledge graphs have proven successful in integrating heterogeneous data across various domains. However, there remains a noticeable dearth of research on their seamless integration among heterogeneous recommender systems, despite knowledge graph-based recommender systems garnering extensive research attention. This study aims to fill this gap by proposing RecKG, a standardized knowledge graph for recommender systems. RecKG ensures the consistent representation of entities across different datasets, accommodating diverse attribute types for effective data integration. Through a meticulous examination of various recommender system datasets, we select attributes for RecKG, ensuring standardized formatting through consistent naming conventions. By these characteristics, RecKG can seamlessly integrate heterogeneous data sources, enabling the discovery of additional semantic information within the integrated knowledge graph. We apply RecKG to standardize real-world datasets, subsequently developing an application for RecKG using a graph database. Finally, we validate RecKG's achievement in interoperability through a qualitative evaluation between RecKG and other studies.

We introduce "causal pieces", a novel concept for analysing spiking neural networks (SNNs), inspired by "linear pieces" used to study expressivity and trainability in artificial neural networks (ANNs). Causal pieces partition the input and parameter space of a feedforward SNN with single-spike coding into distinct regions where the same subnetwork causes the output spikes. For networks of current-based leaky integrate-and-fire (LIF) neurons with large membrane time constants, we show that within each causal piece, output spike times are locally Lipschitz continuous with respect to inputs and network parameters. We further prove a lower bound on the approximation error that depends on the number of causal pieces. Thus, the number of causal pieces is a measure of the approximation capabilities of SNNs, which is valid despite spike-time discontinuities and applies to networks with both excitatory and inhibitory synapses. Empirically, we find that parameter initialisations yielding more causal pieces on the training set strongly correlate with SNN training success across multiple benchmarks, including Yin-Yang, Fashion-MNIST, and EuroSAT. Moreover, simulations with standard single-spike LIF neurons indicate that our findings extend beyond the theoretically analysed regime. These results establish causal pieces as a powerful and principled tool for analysing and improving the computational capabilities of SNNs.

Prompt injection attacks, where untrusted data contains an injected prompt to manipulate the system, have been listed as the top security threat to AI agents. By fine-tuning on simulated prompt injections, SecAlign, a leading open defense, reports LLMs with good test-time robustness and negligible benign utility drop. By scaling up training and evaluations, however, we find that SecAlign actually suffers from significant utility degradation, especially in agentic tasks where the threat of prompt injection is prominent. Motivated by this, we propose Meta-SecAlign for utility-preserving defense by (1) randomized injection position during training to avoid shortcut learning and (2) self-generated responses as high-quality in-distribution training labels. Across general knowledge, instruction following, and agentic workflows (on tool-calling and web-navigation), Meta-SecAlign maintains almost all the undefended LLM's utility while achieving better overall security than SecAlign against various static and GCG adaptive attacks. Experiments use Llama-3.1-8B, Llama-3.3-70B, Llama-4-Scout, Qwen3-4B, and Qwen3.6-27B on 6 prompt injection benchmarks including AgentDojo, InjecAgent, WASP, and SEP. Below are links for the code (https://github.com/facebookresearch/Meta_SecAlign), Meta-SecAlign-70B (https://huggingface.co/facebook/Meta-SecAlign-70B), and Meta-SecAlign-8B (https://huggingface.co/facebook/Meta-SecAlign-8B) models.

Supervised fine-tuning performance for large language models depends strongly on how training budget is distributed across a heterogeneous set of tasks. In practice, mixtures are often fixed using simple heuristics (e.g., uniform or size-proportional sampling) that ignore task interactions, which can hurt transfer and waste budget on redundant sources. We introduce TaskPGM, a framework for learning continuous task mixtures via an energy-based model over tasks. Tasks form the nodes of a Markov random field: unary potentials capture per-task utility, and pairwise potentials encode inter-task relationships using behavioral divergences computed from predictive distributions of single-task fine-tuned models (e.g., Jensen--Shannon divergence and pointwise mutual information). Optimizing this objective yields mixtures that balance coverage against redundancy. We show that the resulting set function is weakly submodular under budget constraints, enabling approximation guarantees for discrete selection variants. Across multiple model families (LLaMA-7B, Qwen2-7B) and evaluation suites (BIG-Bench Hard), TaskPGM improves over standard mixing strategies and provides interpretable structure over task interactions.

We study offline scheduling for large language model (LLM) serving under a fixed KV-cache memory budget, where requests have heterogeneous prompt (prefill) and response (decode) lengths. Given a backlog of requests available at time zero, the scheduler forms mixed prefill/decode batches over time to minimize total end-to-end latency. We show that heterogeneity in prompt lengths fundamentally changes the problem: minimizing total latency is NP-hard, and standard policies that prioritize short outputs or small total sequence sizes can have unbounded approximation ratios. We propose Sorted-F, which repeatedly selects feasible batches using an F-metric that balances batch cardinality against downstream decode cost. With exact batch selection, Sorted-F achieves a constant-factor approximation guarantee in the unit-time, uninterrupted-decoding model with known output lengths; the guarantee also holds under a static peak-memory batch constraint. We develop an exact pseudopolynomial dynamic program for this static subproblem, scalable local-search and greedy heuristics, LP-guided variants, and a receding-horizon online extension. Experiments on public conversational and long-document summarization workloads show that F-metric-based scheduling substantially reduces latency relative to standard baselines and remains close to the LP relaxation lower bound on tractable instances.

Machine learning-based cybersecurity systems are highly vulnerable to adversarial attacks, while Generative Adversarial Networks (GANs) act as both powerful attack enablers and promising defenses. This survey systematically reviews GAN-based adversarial defenses in cybersecurity (2021--August 31, 2025), consolidating recent progress, identifying gaps, and outlining future directions. Using a PRISMA-compliant systematic literature review protocol, we searched five major digital libraries. From 829 initial records, 185 peer-reviewed studies were retained and synthesized through quantitative trend analysis and thematic taxonomy development. We introduce a four-dimensional taxonomy spanning defensive function, GAN architecture, cybersecurity domain, and adversarial threat model. GANs improve detection accuracy, robustness, and data utility across network intrusion detection, malware analysis, and IoT security. Notable advances include WGAN-GP for stable training, CGANs for targeted synthesis, and hybrid GAN models for improved resilience. Yet, persistent challenges remain such as instability in training, lack of standardized benchmarks, high computational cost, and limited explainability. GAN-based defenses demonstrate strong potential but require advances in stable architectures, benchmarking, transparency, and deployment. We propose a roadmap emphasizing hybrid models, unified evaluation, real-world integration, and defenses against emerging threats such as LLM-driven cyberattacks. This survey establishes the foundation for scalable, trustworthy, and adaptive GAN-powered defenses.

In-context learning is governed by both temporal and semantic relationships, shaping how Large Language Models (LLMs) retrieve contextual information. Analogous to human episodic memory, where the retrieval of specific events is enabled by separating events that happened at different times, this work probes the ability of various pretrained LLMs, including transformer and state-space models, to differentiate and retrieve temporally separated events. Specifically, we prompted models with sequences containing multiple presentations of the same token, which reappears at the sequence end. By fixing the positions of these repeated tokens and permuting all others, we removed semantic confounds and isolated temporal effects on next-token prediction. Across diverse sequences, models consistently placed the highest probabilities on tokens following a repeated token, but with a notable bias for those nearest the beginning or end of the input. An ablation experiment linked this phenomenon in transformers to induction heads. Extending the analysis to unique semantic contexts with partial overlap further demonstrated that memories embedded in the middle of a prompt are retrieved less reliably. Despite architectural differences, state-space and transformer models showed comparable temporal biases. Our findings deepen the understanding of temporal biases in in-context learning and offer an illustration of how these biases can enable temporal separation and episodic retrieval.

Temporal reasoning over evolving semi-structured tables poses a challenge to current QA systems. We propose an approach that recasts the task as automated knowledge base construction: (1) prompting an LLM to synthesize a 3NF-compliant relational schema from Wikipedia infobox timelines, (2) populating the schema to obtain a queryable database, and (3) generating and executing SQL queries against it, with QA accuracy serving as an extrinsic evaluation of the constructed knowledge base. In a controlled grid of three schema generators crossed with six query models, the schema source accounts for 79.5% of the exact match (EM) variance against 1.6% for the query model: replacing the schema, and the prompt scaffolding derived from it, shifts EM by 14.7 to 20.0 points, whereas replacing the query model under a fixed schema shifts it by 4.4 to 12.1. From this evidence, we distill three candidate schema-design principles: balanced normalization, semantic naming, and consistent temporal anchoring, framed as correlational hypotheses. Our best configuration (Gemini 2.5 Flash schemas + Gemini-2.0-Flash queries) reaches 80.39 EM, 11.5 points above the strongest reported baseline (68.89 EM); an open-weights configuration reaches 79.52.

Many self-attention sublayers in large language models (LLMs) can be removed with little to no loss. We attribute this to the Attention Suppression Hypothesis: during pre-training, some deep attention layers learn to mute their own contribution, leaving the residual stream and the MLP to carry the representation. We propose Gate-Norm, a one-shot, weight-only criterion that ranks attention sublayers by query-key coupling and removes the least coupled ones, requiring no calibration data, no forward passes, no fine-tuning, and no specialized kernels. On 40-layer, 13B-parameter LLaMA models, Gate-Norm prunes the model in under a second. Pruning 8-16 attention sublayers yields up to $1.30\times$ higher inference throughput while keeping average zero-shot accuracy within 1.5 percentage points of the unpruned baseline across BoolQ, RTE, HellaSwag, WinoGrande, ARC-Easy/Challenge, and OpenBookQA. Across these settings, Gate-Norm matches data-driven pruning methods in accuracy while being $\sim 1000\times$ faster to score layers, enabling practical, data-free compression of LLMs.

Aligning large language models (LLMs) to serve users with heterogeneous and potentially conflicting preferences is a central challenge for personalized and trustworthy AI. We formalize an ideal notion of universal alignment through test-time scaling: for each prompt, the model produces $k\ge 1$ candidate responses and a user selects their preferred one. We introduce $(k,f(k))$-robust alignment, which requires the $k$-output model to have win rate $f(k)$ against any other single-output model, and asymptotic universal alignment (U-alignment), which requires $f(k)\to 1$ as $k\to\infty$. Our main result characterizes the optimal convergence rate: there exists a family of single-output policies whose $k$-sample product policies achieve U-alignment at rate $f(k)=\frac{k}{k+1}$, and no method can achieve a faster rate in general. We show that popular post-training methods, including Nash learning from human feedback (NLHF), can fundamentally underutilize the benefits of test-time scaling. Even though NLHF is optimal for $k=1$, sampling from the resulting (often deterministic) policy cannot guarantee win rates above $\tfrac{1}{2}$ except for an arbitrarily small slack. This stems from a lack of output diversity: existing alignment methods can collapse to a single majority-preferred response, making additional samples redundant. In contrast, our approach preserves output diversity and achieves the optimal test-time scaling rate. In particular, we propose a family of symmetric multi-player alignment games and prove that any symmetric Nash equilibrium policy of the $(k+1)$-player alignment game achieves the optimal $(k,\frac{k}{k+1})$-robust alignment. Finally, we provide theoretical convergence guarantees for self-play learning dynamics in these games and extend the framework to opponents that also generate multiple responses.

Causal inference from observational data can provide strong evidence for finding the best action in a decision-making scenario without having to perform expensive randomized trials. The causal effect of an action is often not pointwise identifiable even with infinite data due to unobserved confounding factors. Furthermore, having only finitely many samples adds another layer of uncertainty to causal effect estimation. Several existing methods can be used to obtain upper and lower bounds to the causal effect, ranging from symbolic methods to the more recent neural network-based approaches, which implicitly incorporate both sources of uncertainty. However, these methods do not inform whether collecting more samples may or may not help identify the best action from observational data, leaving experts in the dark about their data collection strategies. We address this problem with a novel framework that can distinguish the range of causal effect values that might be eliminated by collecting more samples from the range of values that, with high probability, cannot be eliminated with more observational samples. We show that this partitioning can be obtained by solving max-min and min-max optimization problems. We leverage neural causal models to approximately recover this decomposition in practice. We demonstrate via experiments on synthetic and real-world datasets that our algorithm can determine when collecting more samples will not help determine the best action. Our framework can help practitioners decide when to resort to non-observational studies or seek to measure some of the unmeasured confounders for optimal decision-making.

A widespread practice in software development is to tailor coding agents to repositories using context files, such as AGENTS.md. Although this practice is strongly encouraged by agent developers, there is currently no rigorous investigation into whether such context files are actually effective for real-world tasks. In this work, we study this question and evaluate coding agents' task completion performance in two complementary settings: established SWE-bench tasks from popular repositories, with LLM-generated context files, and a novel collection of issues from repositories containing developer-committed context files. Surprisingly, we find that providing context files does not generally improve task success rates, while increasing inference cost by over 20% on average. This observation holds across different LLMs, coding agents, and for both LLM-generated and developer-committed context files. Specifically, we find that while instructions in the context files are well followed by coding agents, repository overviews, although popular and recommended by model providers, are not helpful. We conclude that while context files are useful for specifying non-standard coding practices, any attempts to improve performance should be rigorously evaluated before deployment.

Accurate confidence estimation is critical for reliable use of large language models (LLMs). Prior work on LLM calibration largely focuses on response-level confidence, which estimates the correctness of a single generated output. However, this formulation is misaligned with many practical settings where the central question is how likely a model is to solve a query overall. We show that this mismatch results from the stochastic nature of modern LLM decoding, under which single-response correctness fails to reflect underlying model capability. To address this issue, we introduce capability calibration, a new evaluation framework for measuring how well query-level confidence aligns with a model's expected accuracy on individual queries. We formally distinguish capability calibration (CC) from response calibration (RC) and show that the two differ both theoretically and empirically. We further show that CC is better suited than RC to applications like pass@k prediction and inference budget allocation. Finally, we evaluate common confidence estimation methods to understand the practical feasibility of CC.

We present the first motion generation system for playtesting virtual reality (VR) games. Our player model generates VR headset and handheld controller movements from in-game object arrangements, guided by style exemplars and aligned to maximize simulated gameplay score. We train on the large BOXRR-23 dataset and apply our framework on the popular VR game Beat Saber. The resulting model Robo-Saber produces skilled gameplay and captures diverse player behaviors, mirroring the skill levels and movement patterns specified by input style exemplars. Robo-Saber demonstrates promise in synthesizing rich gameplay data for predictive applications and enabling a physics-based whole-body VR playtesting agent.

We present AgentOptics, an agentic AI framework for high-fidelity, autonomous optical system control built on the Model Context Protocol (MCP). AgentOptics interprets natural language tasks and executes protocol-compliant actions on heterogeneous optical devices through a structured tool abstraction layer. We implement 64 standardized MCP tools across 8 representative optical devices and construct a 410-task benchmark to evaluate request understanding, role-aware responses, multi-step coordination, robustness to linguistic variation, and error handling. We assess two deployment configurations--commercial online LLMs and locally hosted open-source LLMs--and compare them with LLM-based code generation baselines. AgentOptics achieves 87.7%--99.0% average task success rates, significantly outperforming code-generation approaches, which reach up to 50% success. We further demonstrate broader applicability through five case studies extending beyond device-level control to system orchestration, monitoring, and closed-loop optimization. These include DWDM link provisioning and coordinated monitoring of coherent 400 GbE and analog radio-over-fiber (ARoF) channels; autonomous characterization and bias optimization of a wideband ARoF link carrying 5G fronthaul traffic; multi-span channel provisioning with launch power optimization; closed-loop fiber polarization stabilization; and distributed acoustic sensing (DAS)-based fiber monitoring with LLM-assisted event detection. These results establish AgentOptics as a scalable, robust paradigm for autonomous control and orchestration of heterogeneous optical systems.

Reinforcement learning explores effectively in domains such as Atari games, navigation, and locomotion, where novelty over states or dynamics is a sufficient signal. In contrast, dexterous manipulation requires rich physical hand--object interactions, but existing methods often suffer from unstable contact-based novelty signals, inefficient distance novelty signals, or reliance on task-specific priors. We propose ContactExplorer, a general exploration method for dexterous manipulation tasks. ContactExplorer represents contact as the intersection between object surface points and hand keypoints, encouraging dexterous hands to discover diverse and novel contact patterns, namely which fingers contact which object regions. It maintains a contact counter conditioned on discretized object states obtained via learned hash codes. This counter is leveraged in two complementary ways: (1) a count-based contact coverage reward that promotes exploration of novel contact patterns, and (2) an energy-based reaching reward that guides the agent toward under-explored contact regions. We evaluate ContactExplorer on seven contact-rich manipulation tasks and five dexterous hand embodiments. Experimental results show that ContactExplorer substantially improves sample efficiency and success rates over existing exploration methods, that it reduces the need for task-specific priors, and that it remains effective across hand embodiments and transfers to the real world. Project page is https://contact-explorer.github.io.

Vision-language-action policies map camera images and natural-language instructions to a robot's motor actions. Some of these policies are designed to reason in text before acting, generating a reasoning chain and decoding actions conditioned on that chain. The works introducing this design offer the reasoning chain as an oversight interface: text a person can read and edit to correct the policy. What an edited reasoning chain does to the policy's motor actions, whether it repairs them or corrupts them, remains an open question. We measure both directions, repair and corruption, with our deterministic entity swap applied to the instruction the policy receives and to the reasoning chain it generates. A forty-task observed backdrop across all four LIBERO simulation suites reveals that the cost of corrupting the reasoning chain concentrates where language alone determines the goal. There, on LIBERO-Goal, we run the decisive counterfactual intervention with DeepThinkVLA, chosen because its reasoning chain is exposed as plain text. The policy receives a corrupted instruction, paired with the reasoning chain it generates when that instruction is clean. This counterfactually correct reasoning chain recovers 47.8 pp of the lost success, our pre-registered confirmatory test. Had the chain merely restated what the camera image already determines, the injection could have changed nothing. Instead, all 10 tasks move in the predicted direction. The reasoning chain is therefore a working control surface: text written into it steers the robot, repairing behaviour when the text is right and corrupting it when the text is wrong. Whether to expose such a control surface is a real deployment tradeoff, and it can now be measured.

Data-free knowledge distillation enables model compression without original training data, critical for privacy-sensitive tabular domains. However, existing methods does not perform well on tabular data because they do not explicitly address feature interactions, the fundamental way tabular models encode predictive knowledge. We identify interaction diversity, systematic coverage of feature combinations, as an essential requirement for effective tabular distillation. To operationalize this insight, we propose TabKD, which learns adaptive feature bins aligned with teacher decision boundaries, then generates synthetic queries that maximize pairwise interaction coverage. Across 4 benchmark datasets and 4 teacher architectures, TabKD achieves highest student-teacher agreement in 14 out of 16 configurations, outperforming 5 state-of-the-art baselines. We further show that interaction coverage strongly correlates with distillation quality, validating our core hypothesis. Our work establishes interaction-focused exploration as a principled framework for tabular model extraction.

Groundwater represents a key element of the water cycle, yet it exhibits complex and context-dependent relationships that make its modeling challenging. Theory-based models have been the cornerstone of scientific understanding. However, their computational cost, simplifying assumptions, and calibration requirements limit their use. In recent years, data-driven models have emerged as powerful alternatives. In particular, deep learning has proven to be a promising approach for its design flexibility and ability to learn complex relationships directly from the data without requiring extensive domain information. We proposed an attention-based pure deep learning model, named STAINet, to predict weekly groundwater levels in Piedmont (Italy), leveraging both irregular groundwater time series and weather image sequences. To enhance the model's trustworthiness and generalization ability, we merged the theory and data-driven approaches by considering physics-guided strategies to inject the groundwater flow equation into the model. Firstly, we restructured the tail of the architecture to predict the three terms of the governing equation, named the autoregressive, diffusion, and residual components - we thus obtained the PSTAINet-IB. Then, we further injected physics priors by adding loss terms related to the estimated equation components, obtaining the PSTAINet-ILB model. Lastly, we developed the PSTAINet-ILRB by imposing a loss term specific to the residual component, which forces the groundwater recharge to occur within the groundwater body recharge zone, which is identified by domain experts. The models were evaluated both by feeding true lagged values as input and by iterating their own predictions (rollouts) over the whole test set. The PSTAINet-ILB model performed the best, achieving remarkable test performance, and generating equation components in line with domain experts' expectations.

We call query--key relevance absolute when its values lie on a fixed bounded scale, depend on neither competing keys nor sequence length, require no sequence-length-dependent calibration, and can all be zero. To realize this notion, we introduce screening, whose explicit threshold transforms bounded query--key similarities into relevance values, enabling exact rejection, empty selection, and direct inspection on a common scale. In a controlled comparison of 12 attention mechanisms on a matched Transformer backbone, only screening maintains both low long-context perplexity and robust retrieval beyond the training context; notably, it does so without inference-time scaling. Building on screening, we introduce Multiscreen, a language-model architecture composed of parallel gated screening tiles. Multiscreen retains these long-context gains while achieving greater parameter efficiency, stronger general zero-shot downstream performance, lower training cost at larger scales, and lower model-side time to first token than Transformer baselines. We further develop a normalization design that keeps Multiscreen training stable even at a learning rate of $1$ and show that an adapted version likewise stabilizes Transformer at the same learning rate.

Chain-of-thought (CoT) distillation transfers reasoning behaviors from a strong teacher to a smaller student, but prior work reports a capacity gap: distillation may fail when the teacher-student capability mismatch is large. We revisit the capacity gap from a practical perspective by re-examining commonly used experimental settings. Notably, we find that CoT distillation often degrades performance compared to the student's pre-distillation baseline, and that some settings used in prior work, while suitable for establishing the capacity gap as a phenomenon, do not reflect realistic deployment scenarios. Complementing prior work that establishes the capacity gap, we evaluate its practical impact under more realistic settings and find that it does not consistently dominate; stronger teachers tend to be preferable when candidate teachers differ substantially in performance. Our results offer practical guidance for selecting teacher-student pairs in CoT distillation.

Diffusion models generate samples by iteratively querying learned score estimates. A rapidly growing literature focuses on accelerating sampling by minimizing the number of score evaluations, yet the information-theoretic limits of such acceleration remain unclear. In this work, we establish the first score query lower bounds for diffusion sampling. We prove that for $d$-dimensional distributions, given access to score estimates with polynomial accuracy $\varepsilon=d^{-O(1)}$ (in any $L^p$ sense), any sampling algorithm requires $\widetilde{\Omega}(\sqrt{d})$ adaptive score queries. In particular, our proof shows that, within any polynomial total-query budget, successful sampling requires searching over $\widetilde{\Omega}(\sqrt{d})$ distinct noise levels, providing a formal explanation for why multiscale noise schedules are necessary in practice.

Large Language Models are increasingly deployed as interacting agents in settings such as online platforms, recommendation systems, and multi-agent applications. Understanding the collective behaviors that emerge from their interactions is therefore increasingly crucial, especially as these behaviors may shape public opinion and contribute to polarization. In this work, we investigate how network structure and group composition shape the evolution of opinions in populations of LLM agents engaged in multi-round debates. We generate networks with controlled levels of homophily and varying group sizes, and perform ten independent runs per configuration. The results show that LLM agents exhibit patterns that are highly sensitive to network structure, relative group sizes, and to the model itself. We highlight that even within LLM populations, low homophily facilitates convergence by increasing opportunities for cross-group interaction, whereas higher homophily limits cross-group exposure and can preserve distinct opinion states, consistent with previous research. We further show that the choice of LLM substantially affects the resulting dynamics, with different models exhibiting different patterns of opinion updating under comparable network conditions. At the individual level, providing agents with information about their local neighborhood further modulates opinion transitions, revealing substantial differences between models in their sensitivity to local social context. Overall, our findings show that the collective dynamics of LLM populations arise from interactions among model-specific behavior, network structure, population composition, and local social information.

LLM-guided search explores multiple candidate trajectories, but at substantial test-time cost. Pruning low-scoring frontier candidates can control this cost, yet it also turns potentially biased evaluator scores into irreversible decisions: systematic ranking errors can persist under repeated scoring and remove useful branches. We propose Probably Approximately Correct Conformal Filtering (PAC-CF). Its fixed-frontier analysis formulates elimination as an $(\varepsilon,\delta)$-PAC problem under bounded evaluator bias; its operational rule separately calibrates a score-gap threshold on held-out tasks by running the original controller without PAC-CF and using post-search verifier labels to measure the deficit of solution-preserving candidates relative to the frontier leader. Conditional on exchangeable native-controller tasks with nonempty protected exposure, conformal calibration gives finite-sample coverage for retaining at least one verifier-defined valid continuation at every protected frontier on the native trajectory. At deployment, PAC-CF removes only candidates whose gap from the highest frontier score exceeds the frozen threshold. We evaluate PAC-CF across three domains, five controllers, and four request budgets from B100 to B500. In the cross-domain/controller macro averages, the point estimates for all three workload measures are lower at every budget; the paired-bootstrap 95\% confidence interval for utility excludes zero at B100 and B200. For pruning-aware ToolTree, the full-test-set cross-domain utility difference is $+4.38$ points at each tested budget; on the natural-termination sensitivity cohort, physical requests decrease by $18.94$--$18.95\%$ and end-to-end token usage by $23.57$--$23.76\%$.

We propose a novel amortized optimization method for predicting optimal transport (OT) plans across multiple pairs of measures by leveraging Kantorovich potentials derived from sliced OT. We introduce two amortization strategies: regression-based amortization (RA-OT) and objective-based amortization (OA-OT). In RA-OT, we formulate a functional regression model that treats Kantorovich potentials from the original OT problem as responses and those obtained from sliced OT as predictors, and estimate these models via least-squares methods. In OA-OT, we estimate the parameters of the functional model by optimizing the Kantorovich dual objective. In both approaches, the predicted OT plan is subsequently recovered from the estimated potentials. As amortized OT methods, both RA-OT and OA-OT enable efficient solutions to repeated OT problems across different measure pairs by reusing information learned from prior instances to rapidly approximate new solutions. Moreover, by exploiting the structure provided by sliced OT, the proposed models are more parsimonious, independent of specific structures of the measures, such as the number of atoms in the discrete case, while achieving high accuracy. We demonstrate the effectiveness of our approaches on tasks including MNIST digit transport, color transfer, supply-demand transportation on spherical data, and mini-batch OT conditional flow matching.

Agentic workflows carry out complex tasks by orchestrating multiple large language models (LLMs) and tools. Serving them at a target throughput with low latency is hard because they are written in arbitrary agentic frameworks and their execution times are unpredictable: execution branches, fans out, or recurs in data-dependent ways. Since their LLMs often outnumber the available GPUs, they also oversubscribe GPUs. We describe Scepsy, a serving system that schedules arbitrary multi-LLM agentic workflows onto a GPU cluster. Scepsy exploits the insight that, while the end-to-end latency of an agentic workflow is unpredictable, each LLM's fraction of execution time is comparatively stable across requests. Scepsy profiles each LLM under different parallelism degrees and combines the profiles with these fractions into an Aggregate LLM Pipeline, a lightweight throughput and latency predictor for allocations. To minimize latency at a target throughput, Scepsy uses the Aggregate LLM Pipeline to search over fractional GPU shares, tensor parallelism degrees, and replica counts. A hierarchical heuristic then places the chosen allocation onto the cluster, minimizing fragmentation and respecting network topology. On realistic agentic workflows, Scepsy achieves up to 2.5x higher throughput before saturation and 1.0-3.3x lower latency than systems that optimize LLMs independently or rely on user-specified allocations.

Deep reinforcement learning for quantum circuit optimization faces three bottlenecks: replay buffers that overlook temporal difference (TD) target reliability, curriculum-based architecture search requiring a full quantum-classical evaluation after every edit, and the discard of noiseless trajectories when retraining under hardware noise. We address these limitations by treating replay as a central algorithmic lever. We introduce ReaPER+, an annealed replay rule that transitions from TD-error prioritization to reliability-aware sampling as value estimates mature. ReaPER+ achieves up to 4x higher sample efficiency than fixed PER, ReaPER, and uniform replay, while matching prior on-policy solution quality with up to 32x fewer interactions At 12 qubits, fixed ReaPER reaches the lowest energy error in the fewest steps, while PER and uniform replay find more compact circuits at higher error. On tasks scaling to 20 qubits, ReaPER+ retains its advantage, demonstrating that reliability-aware annealing extends beyond small-system benchmarks. LunarLander-v3 confirms that the ReaPER+ is domain-agnostic, it improves success rates by up to 26.8% over PER and 21.8% over fixed ReaPER, with a 3% AUC gain over both. We further introduce OptCRLQAS, which amortizes quantum-classical evaluations across multiple architectural edits, reducing training wall-clock time by up to 67.5% on 12-qubit without degrading solution quality. Finally, lightweight replay-buffer transfer warm-starts noisy optimization from noiseless trajectories, without weight transfer or $\epsilon$-greedy pretraining, reducing steps to chemical accuracy by 85-90% and final energy error by up to 90% relative to from-scratch learning. Transfer gains increase with system size. Together, these results establish experience storage, sampling, and transfer as decisive levers for sample efficient, noise-aware quantum circuit optimization.

Traditional ``row-count $\times$ quality coefficient'' approaches fail to capture the nonlinear utility of data for Large Language Model (LLM) capabilities. We present a utility-aware data valuation framework that moves from static accounting toward utility-based pricing. The framework combines token-level information density and data quality, empirical training contribution estimated by influence functions, proxy models, and Data Shapley values, and cryptographic mechanisms based on hash commitments, Merkle trees, and tamper-evident training ledgers. Preliminary evaluation includes a smoke-scale multi-domain study and a synthetic duplication test. Results show that proxy-based empirical gain achieves strong ranking agreement with proxy-defined realized utility and substantially outperforms row-count and token-count baselines in the tested setting, while the prototype provides a controlled sanity check of the valuation pipeline and its sensitivity to duplication-based manipulation.

In multi-turn program correction, the state-of-the-art method Scattered Forest Search (SFS) has been proposed, employing Monte Carlo Tree Search (MCTS) with carefully crafted initial seeds and text-based optimization. However, since SFS integrates multiple components, the effects of each component on performance and the overall behavior of SFS have not been sufficiently analyzed. In this work, we theoretically analyze the refinement process of SFS from the perspective of version spaces in learning theory and clarify its behavior. First, as a basis for the theoretical analysis, we introduce a sequential self-refinement method (Line), which starts from an initial program and repeatedly refines the resulting program. Furthermore, while Line progresses the refinement process in the depth direction, we introduce Iterative Refinement of Repair Instructions (IRRI) to capture the refinement process in the width direction, which fixes initial programs and iteratively refines repair instructions. We then analyze Line and IRRI and conduct a theoretical analysis of SFS by positioning it as an intermediate method between the two. Our theoretical analysis reveals that SFS exhibits behavior closer to IRRI than to Line, and this theoretical characteristic is also confirmed experimentally. Considering computational resources and methodological simplicity, these results suggest that a simpler method, IRRI, may achieve a similar refinement process without relying on the complex correction mechanism of SFS.

People increasingly turn to general-purpose AI chatbots for advice about emotional and mental health problems, but the ability of these systems to recognize and appropriately triage psychiatric emergencies remains under-characterized. We evaluated psychiatric triage performance in 15 frontier AI chatbots using 112 clinical vignettes spanning four urgency levels, from routine care to immediate emergency assessment. In each trial (1680 total), a chatbot received a single user message conveying all triage-relevant information from one vignette and recommended a timeframe for care. The primary outcome was emergency under-triage; secondary outcomes included triage accuracy and the direction of errors. Vignettes and user messages were generated using a clinician-verified LLM pipeline. Across 415 emergency trials, 23 were under-triaged (5.5%; 95% CI 1.8-15.9). Overall accuracy, averaged across urgency levels, ranged from 42.0% to 71.8% across chatbots and was lowest for intermediate cases (19.6%; 95% CI 11.7-28.1). Every chatbot showed a net over-triage bias; overall, 763 of 786 incorrect assignments (97.1%) were more urgent than the prespecified triage level. The error pattern was similar when predictions were assessed against clinician ratings: 35 of 430 trials involving vignettes rated as emergencies by at least 75% of clinicians were under-triaged (8.1%). AI chatbots recognized most psychiatric emergencies but still missed clinically important cases and frequently over-triaged less urgent presentations. Further evaluations should examine how triage performance changes when clinically relevant information must be elicited through conversation.

Long-horizon memory systems increasingly improve how evidence is stored and retrieved, yet the generator must still reason over fragments whose cross-session relationships are implicit. We study generation-time memory organization as a distinct design dimension and introduce GRAVITY (Generation-time Relational Anchoring Via Injected Topological MemorY), a host-independent auxiliary memory layer. GRAVITY consolidates raw dialogue into entity profiles, temporal event traces, and cross-session topic summaries, then retrieves and injects query-relevant records through the prompt interface. Across five heterogeneous memory systems on LongMemEval and LoCoMo, it improves every host--benchmark baseline under two distinct LLM configurations. Controlled analyses separate gains from organizing already available evidence and from consolidating information across the full history. Under a matched LightMem pipeline, the entity--event--topic representation reaches 83.9% on LoCoMo, 3.6% above the strongest of six alternative auxiliary representations. These results show that generation-time structure is a portable complement to existing memory retrieval, while its interaction with host evidence depends on the benchmark and host.

Memory systems enable otherwise stateless LLM agents to persist user information across sessions, but also introduce a new attack surface. The Trojan Hippo attack is a class of persistent memory attacks that operates under a more realistic threat model than prior memory poisoning work. The attacker plants a dormant payload into an agent's long-term memory via a single untrusted tool call (e.g., a crafted email), which activates only when the user later discusses sensitive topics such as finance, health, or identity, and exfiltrates high-value personal data to the attacker. While anecdotal demonstrations of such attacks have appeared against deployed systems, no prior work systematically evaluates them across heterogeneous memory architectures and defenses. We introduce Trojan Hippo Bench, a dynamic evaluation framework for persistent memory attacks and memory-layer defenses, comprising two components, (1) an OpenEvolve-based adaptive red-teaming benchmark that stress-tests defenses and memory backends against continuously refined attacks, and (2) a capability-aware security-utility analysis for persistent memory systems, enabling principled reasoning about defense deployment across different usage profiles. Instantiated on an email assistant across four memory backends (explicit tool memory, agentic memory, RAG, and sliding-window context), the undefended Trojan Hippo attack achieves up to 85-100% attack success rate (ASR) against current frontier models from OpenAI and Google, with planted memories successfully activating even after 100 benign sessions. We evaluate four memory-system defenses inspired by basic security principles; they reduce attack success to as low as 0-5%, but at utility costs that vary widely with the task mix. Effective real-world deployment thus remains an open challenge that Trojan Hippo Bench is built to study; we release it to support reuse and extension.

Forward-Forward (FF) training lets each layer learn from a local goodness criterion. In cumulative-goodness variants, later layers can inherit a task that earlier layers have partly separated. We formalize this as layer free-riding: under the softplus FF criterion, the class-discrimination gradient reaching block $d$ decays exponentially with the positive margin accumulated by earlier blocks (Theorem 3.1); a squared-hinge barrier reproduces the pathology. History-free and hardness-gated repairs raise deeper-layer separation by up to $5\times$ on CIFAR-100 and $45\times$ in an 8-block CIFAR-10 model, yet these and a depth-scaled auxiliary term move accuracy by under one percentage point among non-degenerate variants; on Tiny ImageNet, a harder cross-dataset check of the selected configuration, layer health likewise does not track accuracy. A theory-derived attenuation-compensated objective raises deepest-block separation $5.9\times$ at a consistent cost in the FF classifier's own accuracy that a probe trained on the frozen features recovers; in a smaller CIFAR-10 control, most of its separation gain comes from dropping the depth-scaled auxiliary term it replaces. Accuracy invariance is a property of the deployed sum-based scoring rule under redistribution (Proposition 3.2) and, empirically, of the structured trained readouts we test, not of what the network learns: retrained variants disagree on 13-26% of test predictions. Architecture and augmentation matter far more than the non-degenerate training rules studied. Cumulative free-riding is therefore a real and repairable pathology, but for the softplus and squared-hinge objectives, architectures, and datasets we study, it is not the dominant factor limiting accuracy.

Adaptive optimizers such as Adam are standard for training Transformers, but storing gradient first and second moments incurs substantial memory overhead. We introduce PowerStep, a memory-efficient optimizer that achieves coordinate-wise adaptivity without storing second-moment statistics. Motivated by $\ell_p$-norm steepest descent, PowerStep applies a signed-power transform directly to one momentum buffer. We establish a finite-horizon stationarity bound for exact, unregularized updates, with an $O(1/\sqrt{T})$ term and a noise-dependent residual. Experiments on Transformers from 124M to 235B parameters show competitive validation quality while halving $\texttt{fp32}$ optimizer-state memory relative to AdamW. Combined with uniform $\texttt{int8}$ quantization, PowerStep remains numerically stable and reduces optimizer-state memory by $\sim8\times$ compared to $\texttt{fp32}$ AdamW. PowerStep thus provides a simple, memory-efficient alternative for large-scale training.

A single base LLM now comes with dozens of post-training variants, quantized, LoRA-adapted, or distilled, and each has to be checked before release. Existing evaluations provide only a partial picture: benchmark scores and likelihood screens say that a variant has degraded, similarity scores such as CKA and SVCCA say how its features moved, and nothing connects the two. We connect them with one structural fact and one design choice: the prediction head is linear, so feature geometry reaches the loss, and we compare the two feature sets through an orthogonal map, which leaves the geometry being measured unchanged. From these we derive PRISM, a closed-form upper bound on the cross-entropy risk gap between a target model and a proxy variant, and prove that it splits exactly into three measurable axes: scale, shape, and head. Each axis names a failure mode and where to intervene: low-bit quantization distorts shape and, at the lowest bit-widths, also inflates activation scale; quantizing the output embedding inflates the head term, which dominates the bound at high bit-widths. Because the shape term is differentiable, the same geometry becomes a regularizer that curbs catastrophic forgetting more than experience replay. Across two model families and five benchmarks, PRISM ranks quantized and fine-tuned variants from a single forward pass at mean Spearman above 0.8, and eight reference sequences already recover the ranking that the risk gap itself needs over a hundred to reach.

Asynchronous reinforcement learning improves rollout throughput for large language model agents by decoupling sample generation from policy optimization, but it also introduces a critical failure mode for PPO-style off-policy correction. In heterogeneous training systems, the total importance ratio should ideally be decomposed into two semantically distinct factors: a \emph{training--inference discrepancy term} that aligns inference-side and training-side distributions at the same behavior-policy version, and a \emph{policy-staleness term} that constrains the update from the historical policy to the current policy. We show that practical asynchronous pipelines with delayed updates and partial rollouts often lose the required historical training-side logits, or old logits. This missing-old-logit problem entangles discrepancy repair with staleness correction, breaks the intended semantics of decoupled correction, and makes clipping and masking thresholds interact undesirably. To address this issue, we study both exact and approximate correction routes. We propose three exact old-logit acquisition strategies: snapshot-based version tracking, a dedicated old-logit model, and synchronization via partial rollout interruption, and compare their system trade-offs. From the perspective of approximate correction, we focus on preserving the benefits of decoupled correction through a more appropriate approximate policy when exact old logits cannot be recovered at low cost, without incurring extra system overhead. Following this analysis, we adopt a revised PPO-EWMA method, which achieves significant gains in both training speed and optimization performance.

Cross-domain offline reinforcement learning learns a target policy from pre-collected source and target datasets with different dynamics. When target data are scarce, effectively compensating for their limited coverage using source data remains challenging due to the discrepancy between domains. We propose Target-aligned Coverage Expansion (TCE), which leverages source states to generatively realign and expand the limited target support while controlling the generation error induced by this expansion. We further derive a performance gap bound that characterizes the interplay between generation error and source--target dynamics gap, providing theoretical guidance for effective coverage expansion and source utilization. Across diverse cross-domain environments, TCE consistently outperforms state-of-the-art baselines, while further analysis empirically supports the guidance provided by our theoretical findings.

Task allocation in smart manufacturing systems must operate under decentralized decision-making, dynamic workloads, and shared-resource constraints. In circular manufacturing settings, these challenges are further intensified because tasks compete for reusable, capacity-constrained assets, and machine selection also might affect processing energy. Although learning-based approaches have been explored for task allocation, improvements in predictive modeling do not necessarily translate into better allocation outcomes under decentralized negotiation. This work proposes an Edge-AI-driven decentralized task-allocation framework. We develop lightweight decision intelligence deployed at the machine level. It is developed progressively: first, a resource-aware heuristic establishes the decentralized bidding structure; a regression-based Edge-AI formulation then examines learned local bid approximation, and a compact autoencoder-regularized pairwise ranking model finally provides a learned correction to the analytical bid ordering. Each machine evaluates incoming tasks by using its processing capability, queue state, energy characteristics, and a compact signal representing contention over the reusable shared production asset. The framework is assessed using discrete-event simulation in scenarios characterized by high load and dependence on shared resources. Compared to the heuristic, the proposed ranking method increases completed tasks, reduces average tardiness, and lowers the deadline-miss rate, with statistically significant paired differences. Mean energy per completed task is also reduced. The results indicate that effective learning-assisted allocation depends not only on approximating local decision quantities, but also on shaping the relative preferences that determine negotiation outcomes.

Public commercial registries are formally open, yet their practical analysis remains difficult because relevant facts are scattered across millions of records that combine structured metadata, multilingual legal notices, temporal events, and entity aliases. This paper presents a controlled, tool-mediated agentic GraphRAG architecture for auditable natural-language analysis of such registries. The proposed pipeline transforms publications from the Swiss Official Gazette of Commerce into a Neo4j knowledge graph comprising over five million nodes and 4.7 million relationships. It combines deterministic ingestion of structured registry fields, LLM-assisted extraction of latent actors from unstructured notices, and a deterministic identity-resolution layer. An analytical agent operates on this graph through intent routing, restricted graph tools, bounded reflection, and state-machine-guided response synthesis. We evaluate the system using a multi-tier protocol covering answer quality, retrieval behavior, entity resolution, and multi-turn conversational performance. The complete architecture is compared with dense, lexical, and hybrid flat-retrieval baselines and with controlled architectural ablations. On a manually curated benchmark, graph-mediated retrieval increases factual correctness from 0.26 for the strongest flat-retrieval baseline to 0.83 for the complete system, with comparable improvements in relevance and completeness. Ablation results show that bounded reflection improves answer quality while intent routing and LLM-based graph enrichment improve reliability in difficult entity resolution tasks. An exploratory dashboard displays the graph evidence and execution traces underlying each response, allowing users to inspect how answers were produced.

LLM routing matters most in long-horizon applications such as coding agents, deep research systems, and computer-use agents, where a single user request triggers many model calls. Routing each call to the cheapest sufficient model can cut costs without sacrificing quality, yet existing router benchmarks evaluate routers only on one-shot prompts. They never expose the router-visible prefix at an intermediate agent step, never test whether a cheaper replacement preserves downstream task success, and often rely on online LLM judges at evaluation time. We introduce TwinRouterBench, a step-level routing benchmark with two tracks. The static track provides 970 router-visible prefixes from 520 instances across SWE-bench, BFCL, mtRAG, QMSum, and PinchBench, each paired with an execution-verified target tier estimated under a released downgrade-and-cascade protocol; scoring is deterministic arithmetic over tier labels, trajectory membership, and token costs, with no online evaluator-side LLM judge. The dynamic track supplies a harness that runs routers on the full 500-case SWE-bench Verified suite; in this paper we report a 100-case held-out evaluation disjoint from the static SWE supervision split. At each LLM call the router selects a concrete model from a locked pool, and success is measured by official task resolution and realized API spend. The two tracks support fast offline iteration followed by end-to-end validation under live agent execution. Code and data are available at https://github.com/CommonstackAI/TwinRouterBench.

Chunked vision-language-action (VLA) policies generate multi-step actions from one observation and typically re-observe after executing the action chunk. Earlier actions change the scene conditions for later steps, while occlusion can compromise visual feedback. Spatial and temporal VLAs enhance geometry and memory, but their representations may become stale as actions alter object poses and contacts. We introduce EvoScene-VLA, which uses compact scene tokens to unify within-chunk scene prediction, cross-chunk state propagation, and observation-based correction. The action expert jointly denoises actions and future scene states in a single flow-matching process, allowing predicted scene changes to inform action generation. The final scene state serves as the next prior for subsequent correction. At the next control call, the vision-language model (VLM) fuses the incoming visual input with this prior. Two-level geometric anchoring applies local depth supervision to per-view scene features and global 3D feature supervision to the fused scene state; a shared geometric decoder imposes the same constraints on current and future states. Since demonstrations do not directly provide future scene states, we train a scene predictor using 3D teacher features from future images. Its predictions provide targets for the action expert's future scene states. At deployment, we remove both Scene Predictor and Geometric Anchor. EvoScene-VLA improves average success over the matched baseline by 2.4 and 2.2 percentage points on 31 RoboTwin tasks under fixed and randomized settings, respectively, and by 2.7 points on LIBERO. Across three real-robot tasks, average success rises from 37.3% to 42.0%. Ablations show that a recurrent scene state alone does not necessarily improve performance, whereas state propagation can improve control when combined with geometric and future scene supervision.

Frontier language models sometimes recognize that they are under evaluation and adjust their behavior which can undermine validity of benchmark results. Yet the field studies it without a shared foundation, conflating flaws of the evaluation with capabilities of the model, and detection with behavioral response. We ground evaluation awareness in social psychology, decomposing it into an environment component and a model component that separates recognition from propensity. We operationalize the environment component through eight categorized trigger factors, such as placeholder entities and grading-style output formats, and study recognition and behavior through chain-of-thought monitoring. Across nine frontier models and four benchmarks, recognition rates depend on the specific pairing of model and benchmark. Recognition rarely associates with behavioral change, and when it does, the direction depends on the type of evaluation perceived. Models are also more sensitive to safety than capability evaluations, placing safety benchmark validity at greater risk. To study which factors each model is sensitive to and how they interact, we propose \textbf{EvalAwareBench}, a factor-controlled benchmark of 100 paired safety-capability tasks where each of the eight factors can be independently toggled, varying evaluative signals while holding the underlying request fixed. Through EvalAwareBench, we find that no single factor uniformly affects all models, but stacking factors progressively raises evaluation awareness across all of them. Our framework and EvalAwareBench provide the tools to measure, attribute, and mitigate evaluation awareness, building the foundation for future solutions.

Fast and reliable optimal power flow (OPF) approximation is important for power system operation, yet heterogeneous OPF graph models are often evaluated either with architecture-specific implementations or on limited training corpora. This paper presents a large-scale heterogeneous graph-learning workflow, built on HydraGNN, for data-driven OPF surrogate modeling and graph foundation-model (GFM) development. The workflow preserves buses, generators, loads, shunts, alternating-current (AC) lines, transformers, and device-to-bus relations and provides a common implementation for distributed preprocessing, multi-graphics-processing-unit (GPU) training, hyperparameter optimization (HPO), and downstream adaptation. Using approximately three million heterogeneous graphs from ten Power Grid Library for Benchmarking AC Optimal Power Flow Algorithms (PGLib-OPF) cases spanning 14 to 13,659 buses, we compare six heterogeneous graph neural network (GNN) families under a common HPO campaign on Frontier, resulting in two compact HeteroSAGE and HeteroHEAT models (~1.6-1.7 million parameters) that achieve the lowest observed validation MSE. Strong and weak scaling using up to 1,024 Frontier nodes show that 256 nodes (corresponding to 1,024 AMD MI250X GPUs) provides the best time-to-solution balance for the measured workload. Two downstream tasks evaluate task adaptation on the IEEE 118-bus case: a classification task to distinguish nominal operating conditions from artificially generated extreme-overload conditions and N-1 single-component-outage regression to predict bus voltage magnitude and phase angle. Partial fine-tuning provides the strongest balance between predictive performance and computational requirements.

The linear growth of the Key-Value (KV) cache is a critical bottleneck in long-form LLM inference. Existing KV compression methods mitigate this by evicting tokens based on importance scores. However, we show that their reliance on global Top-k selection triggers Region Wipe-out: the severe eviction of contiguous reasoning blocks that derails logical coherence. To address this, we propose Adaptive Mass-Segmented (AMS) KV Compression, a framework that shifts the paradigm from token-level competition to region-aware quota allocation. AMS adaptively partitions the KV cache based on the spatial distribution of attention mass, ensuring structurally vital reasoning segments receive guaranteed memory quotas. To ensure stability during iterative decoding, an EMA-based smoothing mechanism is incorporated to prevent jitter in segment boundaries. Crucially, AMS is a universal plug-and-play layer that is orthogonal to existing scorers. It can be seamlessly integrated into representative methods such as TOVA, Expected Attention, KeyDiff, R-KV and TriAttention. AMS is also system-compatible with modern paged-KV serving frameworks such as vLLM, supporting efficient gather-and-compact KV execution without introducing additional steady-state attention overhead. Extensive experiments across a diverse suite of tasks, including mathematical reasoning (MATH500, AIME, GSM8K), code completion, open-domain QA, and sparse retrieval, demonstrate that AMS consistently mitigates structural fragmentation and boosts model performance.

We address the problem of learning to assign prediction tasks to one agent from a set of available agents, including human decision-makers and AI models. We focus on sequential learning of agent expertise and assignment policies where each agent is constrained to handle a fraction of tasks. We provide a general theoretical characterization of this problem in terms of agent capacities, differences in agent expertise, and task context. We then develop a framework of sequential explore-exploit policy-learning algorithms that seek to maximize overall performance. Experimental results over a variety of tabular, image, and text prediction tasks demonstrate systematic gains from our policy-learning algorithms relative to non-contextual baselines across different types of agents, including LLMs and humans.

LLM pretraining is extremely costly; therefore, parameter-efficient LLM architectures have recently emerged as a compelling research direction. One such promising approach is to represent the parameters as orthonormal low-rank weight matrices. However, maintaining orthonormality during training is computationally expensive, making it impractical. This paper presents the TSVD (Truncated Singular Value Decomposition) framework which efficiently maintains orthonormality through QR decomposition and caching. Furthermore, a spectral energy heuristic is introduced to select the rank of the resulting low-rank weight matrices. Empirical evaluations across model sizes show that TSVD matches or outperforms full-parameter baselines at a fraction of the compute cost. TSVD thus provides a scalable, computationally efficient foundation for LLM pretraining.

Proactive assistants continuously decide when to intervene and what context should support the intervention. Large language model (LLM) pipelines repeatedly interpret activity histories to make these decisions, paying an inference cost even when the assistant remains silent. We show that a small graph model can handle both decisions and improve the language agents it controls. Our key insight is that user activity has a native graph structure: events involve persistent entities whose recurrence connects interactions over time. Triggering and context selection map directly to predictions on event and entity nodes. We introduce a temporal-graph-learning (TGL) controller that learns these predictions jointly and supplies both outputs in one forward pass. The downstream language agent generates suggestions on triggered events using the activity history and scored entities. At 11.13 ms per event on a GPU server, TGL achieves the highest AUCs among nine trigger architectures and gives approximately $4$--$7\times$ trigger-stage speedups over the two single-forward LLM triggers. A shared TGL model improves F1 across all 14 downstream backbones by a mean of 16.7 points. The controller also runs at 13.99 ms on a consumer laptop with an approximately 220 MiB BF16 resident footprint, bringing effective proactive control to on-device deployment.

Frontier reasoning models are produced by post-training base language models with reinforcement learning. Recent work has challenged this by showing that sampling from a sharpened version of the base model's distribution, a so-called power distribution, elicits comparable reasoning without additional training, curated datasets, or verifiers. However, making this method practical requires efficiently sampling from the power distribution. A sampler needs to "mix" to the power distribution, which necessitates moving between modes of the target distribution; intuitively, e.g., trying different reasoning strategies. The samplers proposed in prior works repeatedly select a "cut" position in the current reasoning trace uniformly at random and resample the suffix from that position onward. However, reasoning traces typically contain a few consequential decisions (e.g., the choice of proof strategy or algorithm), and we observe that a uniformly chosen cut tends to rewrite local details rather than revisit decision points. We introduce an algorithm (Entropy-Cut Metropolis-Hastings) that uses the base model's next-token entropy as a proxy to identify key decision points and resample from those positions. We empirically verify that entropy jumps are a useful proxy for decision points and, in a stylized model of reasoning, prove that our method's mixing time scales with the number of decisions in a trace rather than with the number of tokens, which can be much larger. Across MATH500, HumanEval, GPQA Diamond, and AIME26, our method consistently improves over baselines and RL-trained models, including best-of-N at the same token budget.

We study fixed-confidence best-action identification (BAI) in stochastic minimax trees. This problem is increasingly relevant in modern AI planning, where deep minimax search and Monte Carlo Tree Search (MCTS) with language model long rollouts face a fundamental tradeoff: heuristic evaluations are cheap but biased, while accurate rollouts are reliable but prohibitively expensive. We propose 2FFS, a two-fidelity tree-search algorithm that brings multi-fidelity flat bandit ideas into trees. The algorithm combines minimax-style fast expansion with MCTS-style stochastic sampling, adaptively deciding when to exploit cheap biased evaluations and when to invoke expensive accurate evaluations for local certification. We prove fixed-confidence correctness, establish finite stopping for exact identification, and give a polynomial-depth cost upper bound for general-depth trees. Across numerical stochastic-tree experiments, 2FFS uses substantially fewer samples and computational operations comparing to existing BAI-MCTS baseline.

While household robots are often evaluated based on task completion, everyday domestic environments involve value-conflicting situations where robots are expected to choose actions that prioritize diverse values such as human autonomy, efficiency, or social appropriateness. Yet, there are no benchmarks for evaluating robots' value preferences in such scenarios. We introduce RobotValues, a benchmark to evaluate household robot planners in 8K value-conflict scenarios. Each instance consists of a realistic, synthetically generated household image with multiple plausible robot actions that prioritize different human values. We construct ROBOTVALUES through LLM-assisted scenario generation, stakeholder-grounded value extraction, image generation and automatic quality control. We evaluate 10 VLMs used in robotics and find that models show default value preferences, including safety and accommodation, while underselecting privacy-prioritizing actions. When models are prompted to prioritize values that conflict with their preferences, models often fail to override the default actions, choosing incorrect actions 80% of the time on average across models. These findings highlight the need to go beyond task completion or safety evaluations and assess robots' decision-making capability when human values conflict.

Rubric-based reinforcement learning (RL) uses an LLM-as-a-Judge (LaaJ) to score model outputs according to rubrics as rewards. However, policy models may exploit latent biases in the judge, leading to reward hacking and ineffective or unsafe training outcomes. In real-world rubric-based RL, such hacking behaviors are often subtle and entangled with multiple judge biases, making them difficult to analyze, detect, and mitigate. In this paper, we introduce CHERRL, a Controllable Hacking Environment for Rubric-based RL. By injecting known biases into LaaJ, CHERRL enables stable reproduction of reward hacking, explicit observation of reward divergence, and identification of hacking onset. This provides a clean experimental testbed for studying the mechanisms and mitigations of reward hacking in rubric-based RL. To demonstrate its utility, we analyze different judge biases from the perspectives of discoverability and exploitability, and explore an agent for automatically detecting reward hacking onset from training logs. The code and environment are publicly available at https://github.com/THUAIS-Lab/CHERRL.

Humanoid loco-manipulation benefits from one controller to coordinate locomotion, arm movements, and fall recovery. These behaviors are difficult to learn together from scratch because they require different skills and objectives. We present HANDOFF, a training architecture that distills motion-tracking, locomotion, and recovery teachers into one 29-DoF policy, using context-conditioned KL losses over separate action slices and a soft mixture of experts. Commanded velocity continuously blends body supervision, while a binary recovery flag assigns the full action to the recovery teacher. The resulting controller takes a compact 10-D task-space command rather than a dense kinematic stream and does not switch policies at runtime. On the Unitree G1, HANDOFF matches state-of-the-art velocity tracking and obtains the largest robust manipulation workspace in a matched adapted-interface benchmark. The same controller executes natural-language-driven, multi-stage loco-manipulation tasks in simulation and hardware experiments without further data collection or fine-tuning.

In time-series generation, existing approaches typically handcraft ortrain a separate model for each dataset, which hinders their scalability and fails to leverage shared temporal structures across domains. To address this fragmentation, we propose UPLOTS, a Unified, Prompt-guided Language model framework fOr constrained Time-Series Generation across diverse domains. Instead of building task-specific models, UPLOTS leverages a single pre-trained transformer backbone guided by learned constraint prompts, enabling on-demand generation with precise pattern control. One key innovation is our dynamic multi-dataset loss re-weighting and prompt-to-pattern mapping, which allows UPLOTS to internalize diverse temporal structures during training and conditionally generate them at inference. We evaluate UPLOTS on four real-world benchmarks and multiple constraint settings, including peak-period, calendar, load-level, and volatility patterns. Additional held-out constraint-combination and downstream forecasting experiments further demonstrate that UPLOTS generalizes beyond the original peak-pattern setting and improves data augmentation under scarce real-data regimes. Our code and baselines are available at github repo: https://github.com/cruiseresearchgroup/UPLOTS.

Language models trained on large-scale vision-language data have demonstrated strong potential for embodied agents. Harnessing models through embodied tools use offers a promising alternative to end-to-end vision-language-action systems by combining high-level reasoning with external modules for perception, planning, and control. However, it remains unclear what makes an effective harness for embodied manipulation, and to what extent such a harness can unlock embodied capabilities in a wide range of reasoning models. In this work, we present Guava, a harness framework for embodied tool use developed through systematic exploration of the design space of agent workflows, action spaces, and observation spaces. Our study identifies three key ingredients for effective embodied agents: iterative perception-reasoning-action loops, semantic action abstractions, and multimodal observations. To understand whether these design principles are universal even to small models, we develop an end-to-end training pipeline that distills embodied manipulation capabilities into a 4B open-source model using fewer than 2K trajectories collected entirely in simulation. Experimental results in both simulation and real-world environments show performance comparable to frontier proprietary models while exhibiting strong generalization to unseen objects, novel instructions, and long-horizon tasks. Results suggest that a well-designed harness can serve as a scalable, model-agnostic interface for embodied manipulation, enabling strong emergent embodied capabilities in compact open-source models with minimal training data.

Large Language Model (LLM)-based automatic Multi-Agent Systems (MAS) generation has become a crucial frontier for tackling complex tasks. However, existing methods face a dilemma between model capability and experience retention. Inference-time MAS leverages frozen frontier LLMs but repeats identical searches without learning from past experience. Conversely, Training-time MAS internalizes experience via gradient updates but is constrained by the low capability ceiling of smaller models, and is hard to scale to large frontier LLMs. To bridge this gap, we propose Skill-MAS, a novel third path that decouples experience retention from parametric updates by conceptualizing the high-level orchestration capability as an evolvable Meta-Skill. Skill-MAS refines this architectural knowledge through a closed optimization loop: (1) Multi-Trajectory Rollout samples a behavioral distribution for each task under the current Meta-Skill; and (2) Selective Reflection adaptively selects priority tasks and applies hierarchical contrastive analysis to distill systemic experience into generalizable, strategy-level principles. Extensive experiments across four complex benchmarks and four distinct LLMs demonstrate that Skill-MAS not only achieves remarkable performance gains but also maintains a favorable cost-performance trade-off. Further analysis reveals that the evolved Meta-Skills are highly robust and exhibit strong transferability across unseen tasks and different LLMs.

Long-horizon policies trained with reinforcement learning can still achieve high return through inefficient interactions, while rare efficient behaviors discovered during training may be forgotten. We argue that temporal efficiency itself provides a source of self-supervision for reinforcement learning. We introduce Temporal Self-Imitation Learning (TSIL), a reinforcement learning framework that mines temporally efficient successful trajectories generated during learning and converts them into reusable supervision for future policy improvement. TSIL progressively refines learning using configuration-conditioned adaptive temporal targets derived from fast successful trajectories, while preserving and replaying efficient behaviors through efficiency-weighted self-imitation learning. Across 30 long-horizon tasks spanning robot manipulation and interactive navigation, TSIL consistently improves task success rates, learning efficiency, behavioral efficiency, and robustness to unstable training conditions.

Maintainability is a core dimension of software engineering, shaping how code is written, reviewed, and developed over time. While coding agents have demonstrated strong performance on single-issue tasks, it remains unclear how maintainable their code is when future agents build on top of it, potentially leading to compounding downstream effects. We investigate how agent code compares to human code in these maintenance settings, presenting CodeThread, a framework to construct controlled experiments from repository-level coding benchmarks. Applying CodeThread to four frontier coding agents and four benchmarks, we find that agents are less effective at resolving tasks when building on agent code compared to human code, with task resolve rate drops of up to 13.1%. Regression analysis reveals that many traditional software engineering maintainability metrics do not explain this difference. Instead, the clearest signals are subtler behavioral differences in agent code, such as changes to input validation and error handling, along with differences in downstream code size and task difficulty. These findings highlight the need to evaluate these systems not only by immediate task resolution but also by code maintainability, and point to potential sources of downstream errors introduced by agent code.

Physical neural networks promise low-power machine learning by computing directly with analogue device physics, but most architectures force nonlinear device responses to act as scalar weights. Inspired by Kolmogorov-Arnold networks, we place trainable nonlinear functions on the connections, making each physical connection a learnable computational element. Realising these functions as analogue band-pass filters on field-programmable analogue arrays, we find that the benefit is task-dependent and follows from the smoothness of the physical basis: the networks represent smooth, continuously valued targets, including robotic kinematics, continuous control, and photovoltaic maximum-power-point tracking, with far fewer nodes and connections than multilayer perceptrons, but offer no parameter-efficiency advantage on classification-like decision boundaries. Trained networks transfer to hardware across approximately 35,000 connections with quantified fidelity, and a dedicated CMOS implementation is projected to operate at approximately 30 microwatts. A memristive realisation reproduces the same behaviour in simulation, indicating that the advantage comes from placing trainable nonlinearity on connections, rather than from a particular device.

Public discourse about artificial intelligence (AI) often uses anthropomorphic language: language that attributes human capabilities and characteristics to AI systems. This practice has been criticized for setting misleading expectations, inflating claims, and fueling hype around AI, which may distort public understanding of AI and impact policy priorities. We study the effects of anthropomorphic framing by comparing changes in participants' perceptions of AI (N=815) when reading passages with and without anthropomorphic language, designed to reflect realistic public-facing AI discourse. We further examine whether these effects differ across two types of AI technologies -- large language models and recommendation systems -- and measure changes in perceptions of AI across several dimensions that are prominent in current public discourse. In a separate condition using a text that explicitly discusses the dangers of AI, we show that individuals' views of AI can shift in response to reading a text; yet in the main conditions of the experiment, where we compare anthropomorphic and non-anthropomorphic descriptions, we find that whether the text uses anthropomorphic language does not substantially affect participants' perceptions of AI. Our results indicate that any immediate effects on opinions of AI are modest, although they leave open the possibility that anthropomorphic language could have an effect in naturalistic settings, or over gradual, continued exposure.

Understanding which parameters are influential in Large Language Models (LLMs) is central to improving their efficiency, reliability, and interpretability. We introduce Weight-Adjusted Gradients (WAG), a simple yet effective approach for estimating parameter importance that explicitly captures the interaction between model weights and first-order gradient information and identifies parameters that disproportionately influence model behavior, such as those responsible for collapse phenomena in LLMs. Across a range of models and settings, we show that WAG surfaces a tiny but critical subset of parameters (< 0.5 parts per million or 0.00005% of model size) whose modification leads to dramatic degradation in performance, indicating a novel failure mode. These findings also reveal a previously underexplored interplay between weights and gradients, suggesting that parameter importance cannot be fully understood through either signal alone. We demonstrate the practical utility of WAG across several diverse applications, such as expert allocation in Mixture-of-Experts (MoE) architectures, targeted unlearning, mixed-precision quantization, and layer selection for knowledge editing. In sum, WAG can serve as a unified approach for analyzing, debugging, and controlling LLMs, and opens new directions for principled parameter-level interpretation.

Conversational AI can improve access to public health information, but public-facing healthcare applications require safeguards against inappropriate medical guidance and unsupported generation. We present a Safety-First Science Gateway for maternal and child health (MCH) resource navigation that combines large language models (LLMs) and retrieval-augmented generation (RAG) with a multi-layer safety architecture. The gateway integrates emergency handling, domain/scope screening, source attribution, anonymous session management, and operational audit logging while restricting retrieval to curated institutional resources. We describe the gateway architecture, prototype implementation, and functional verification of selected workflows. The current system provides resource provenance and safety-bounded navigation; it does not constitute a clinical decision-support system or automated claim-by-claim verification of generated health information. This work provides a reusable architectural framework for conversational navigation of curated public-health resources.

Large language model (LLM) post-training is essential for improving reasoning, adaptation, and alignment. Existing methods mainly follow two paradigms: reinforcement learning (RL) and on-policy distillation (OPD). However, RL relies on coarse-grained outcome supervision, resulting in difficult credit assignment and limited capability to acquire new knowledge. OPD, meanwhile, unconditionally matches teacher logits through KL divergence, which creates a dilemma: similar teachers provide little new knowledge, while substantially different teachers often yield ineffective guidance, largely restricting OPD to within-family distillation. We propose Distilled Reinforcement Learning (Distilled RL), which integrates teacher supervision into the RL objective to provide fine-grained guidance, selectively transfer new knowledge and avoid unconditional imitation. Distilled RL contains three components: reverse importance sampling with clipping, negative sample reset, and sequence-level geometric normalization. Through a concise and interpretable case study, we demonstrate that Distilled RL can effectively transfer previously unavailable knowledge from a teacher model to a student model. Extensive experiments across both within-family and cross-family distillation settings show that Distilled RL substantially outperforms standard RL and OPD in terms of both pass@1 and pass@k. Our code is available at https://github.com/597358816/Distilled-RL.

Serving large language models at long context is bottlenecked by the key-value (KV) cache, which is read at every decode step. We find that attention keys are approximately low-rank within pages. A single low-rank projection shared across pages can miss page-specific directions; fitting a basis to each page better identifies the pages receiving the most attention at comparable stored selector cost. LOCKS stores a rank-$r$ spectral summary per page, reconstructs its within-page logits, and selects pages by log-sum-exp mass without reading candidate keys or values. It stays within about a point of FullKV on LongBench-v1, tracks the read-every-key exact-LSE oracle on RULER down to the smallest budgets, and retains quality furthest under tight budgets on AIME26 and MATH-500. At a $2048$-token budget it matches FullKV aggregate quality beyond $100$K context while attending about $2\%$ of tokens. Across ranks $2$-$8$, summaries use $4$-$10\%$ of full-KV bytes. On GH200 with GPU-resident KV, LOCKS reduces complete decode-step time by $1.8\times$ at $512$K context. With full KV offloaded to Grace memory, it reaches $3.82$-$4.22\times$ the faster dense backend's aggregate throughput at $64$K-$256$K by serving larger batches.

Weak supervision sets a practical regime for audio-visual sound source localization as dense spatial annotations are costly to obtain at scale. The task, however, remains challenging, as models must locate sound sources from temporally aligned audio-visual data without pixel-level supervision. Recent large-scale audio-visual retrieval models, trained at unprecedented scale, encode rich multimodal structure. We show their latent representations, though optimized for global alignment, can nonetheless enable fine-grained spatial grounding. While spatial detail is progressively lost in the upper layers of retrieval backbones due to global pooling, intermediate visual tokens retain highly structured spatial information. To exploit this, we introduce LAIP (\emph{Localization via Audio-Informed Pooling}), a framework that employs a lightweight \emph{Audio-informed Spatial Pooling} (AiSP) to replace the standard global aggregation module. By querying intermediate visual tokens with audio aligned at the frame level, LAIP recovers localized spatial information that is otherwise discarded by the retrieval pipeline, with the largest gains observed for PE-AV, a stack with underlying temporal aggregation. Our approach achieves state-of-the-art performance on AVSBench and AVATAR, nearly doubling previous results on the latter, improving average CIoU from 13.21 to 26.22.

Large language models (LLMs) have demonstrated strong reasoning capabilities, motivating their use as the backbone of foundation recommendation models (FRMs). Existing methods enhance recommendations through explicit Chain-of-Thought (CoT) reasoning under a Think-then-Answer paradigm. However, explicit CoT incurs substantial inference overhead by generating lengthy reasoning traces and relies on manually designed templates that struggle to capture diverse, dynamic user interests. We propose OneLatent, an efficient latent reasoning framework that compresses explicit reasoning traces into several learnable latent tokens, enabling Latent-Reason-then-Answer inference without generating verbose traces. OneLatent first introduces Multi-View Adaptive CoT (MV-ACoT), which creates diverse, high-quality teacher-generated supervision by exploring user interests from multiple perspectives and automatically adapting reasoning complexity to each instance. Building on pretrained FRMs, it then uses a three-stage latent-token alignment paradigm to progressively internalize CoT traces into learnable latent tokens. Finally, a multistage curriculum-based post-training strategy activates latent-token reasoning for downstream recommendation tasks. Experiments on an industrial-scale Kuaishou dataset and the public Kuaishou LLM-Rec benchmark show that OneLatent consistently outperforms explicit CoT-based methods and traditional baselines. Compared with the Think and No-Think variants of FRMs, OneLatent improves SID@64 by 17.44% and 9.33%, respectively, while achieving over 17x higher online inference throughput. We further develop a production serving system for scalable, real-time FRM inference. An online A/B test in Kuaishou's local-services advertising scenario shows that deploying OneLatent with this system yields an estimated 9.6% revenue lift over strong online baselines, including OneRec and OneReason.

Large language models can write, patch, and search code, but oncall root cause analysis (RCA) demands something different: reasoning over noisy metrics, logs, traces, and source code, starting from ambiguous user-facing reports, often hours after the incident began. We introduce ORCA-bench, a benchmark that puts general-purpose coding agents in a production-fidelity oncall setting. ORCA-bench pairs 1,079 RCA tasks with six days of metrics, logs, and traces collected from an OpenTelemetry-instrumented microservice system under continuous simulated user load. Agents investigate this recorded history through real observability interfaces---Prometheus, Jaeger, and OpenSearch via Grafana---with full access to application source code. Tasks systematically vary report specificity, time-to-detection, and co-occurring fault scenarios. Ground-truth symptoms are curated and signed off by expert SREs, and our LLM-as-judge is independently re-scored by humans (Cohen's $\kappa_w = 0.91$). Across five frontier agents, the best RCA Accuracy is 25.3% on Medium-difficulty tasks (the realistic-input setting) and 10.0% on Hard---a gap that remains even with Claude Fable 5. The weakest model hallucinates an implausible root cause in 40% of incident reports, and removing source-code access reduces RCA accuracy and increases the hallucination rate for every evaluated model. These results come from a curated 50 GB / six-day testbed of standalone tasks on a system whose code and instrumentation are public. Since real production systems are orders of magnitude larger, more dynamic, and more idiosyncratic, the gap we report underscores the engineering work still needed before agents can be entrusted with production reliability. We release the public set at https://hub.harborframework.com/datasets/orca-bench/orca-bench.

We present PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The deployed policy sees the ball only as segmentation-masked depth from a head-mounted camera, while training-time CBF guidance represents clearance to every body link, and an adversarial motion prior regularizes the resulting evasive reflexes. We evaluate on a controlled any-link contact benchmark with seeded throws in two regimes: single throws and a deployment loop in which the robot walks back to its station and recovers between throws. On this benchmark, the policy comes within a few points of a privileged state oracle: a fixed onboard camera alone is adequate for evasion. We find that usable barrier structure depends on perceptual observability: Joint-CBF gives the best performance with accurate ball states, degrades under fixed-camera observations when used only as training guidance, and recovers with a ball-tracking gimbal or privileged runtime filter. We therefore deploy a lightweight Link-CBF policy zero-shot on the Unitree G1 in the real world, where it tolerates imperfect perception, succeeds on 95% of throws, and uses semantic segmentation to dodge different balls.

Domain adaptation of small language models (SLMs) has emerged as a practical strategy for deploying capable NLP systems in resource-constrained, high-stakes environments including healthcare, legal services, and financial analysis. While performance gains from parameter-efficient fine-tuning are well characterised, the corresponding impact on trustworthiness (factual calibration and adversarial robustness) remains poorly understood. This paper presents the first systematic cross-domain, cross-architecture empirical study quantifying the trustworthiness cost of domain adaptation across three SLM architectures (TinyLlama 1B, Gemma-2 2B, Llama 3.2 1B), three domains (healthcare, legal, finance), two training-data conditions (benign and adversarially perturbed), and four fine-tuning strategies (baseline LoRA, Safety-DPO, Dark Experience Replay, and Task Arithmetic LoRA, TA-LoRA). Trustworthiness is evaluated through TruthfulQA MC2 (factual calibration) and HarmBench ASR (adversarial robustness) across all 216 experimental configurations with three random seeds. Three principal findings emerge. First, baseline QLoRA domain adaptation produces minimal TruthfulQA MC2 change across all model-domain combinations (mean |Delta TQA| < 0.02). Second, adversarially perturbed training data consistently improves domain adaptation quality (Delta loss approximately -0.040) without worsening trustworthiness benchmarks. Third, none of the three safety-preserving strategies reduced adversarial harm susceptibility: Safety-DPO was effectively neutral (mean Delta ASR < 0.001), while Dark ER and TA-LoRA increased mean HarmBench ASR by +0.171 and +0.155 respectively in safety-aligned models (Gemma-2 2B, Llama 3.2 1B), with individual configurations exceeding +0.45. These results challenge the assumption that replay-based and arithmetic-merge strategies transfer alignment to domain-adapted SLMs.

Coding agents powered by large language models (LLMs) are increasingly adopted in software engineering (SWE) scenarios, capable of fixing a specific bug in large-scale codebase. However, existing SWE benchmarks typically assume that high-quality issue reports with detailed information are always available, which is easily violated in practice due to the complexity of report acquisition and curation. To address this, we introduce Active-SWE, a benchmark for evaluating coding agents on proactively discovering and fixing multiple bugs without report guidance, covering 1,663 tasks across six bug categories and eight languages. Beyond shifting the focus from existing reactive bug fixing to proactive bug fixing, Active-SWE enables a more in-depth evaluation by expanding the scope from fixing a specific recorded bug to multiple-bug fixing and potential bug discovery scenarios. To construct Active-SWE, we propose a novel difficulty-aware task formulation pipeline with a dual-track evaluation framework, facilitating comprehensive evaluation of proactive bug-fixing capability. Extensive experiments reveal that most state-of-the-art coding agents struggle with proactive bug-fixing tasks, demonstrating limited performance in locating and resolving recorded bugs, handling multiple bug fixing scenarios, and discovering valid potential bugs.

On-policy self-distillation (OPSD) converts feedback into dense token-level supervision on student-generated trajectories, complementing reinforcement learning with sparse outcome rewards. Its self-teacher, derived from the student's current or exponentially averaged parameters and conditioned on additional context, evolves alongside the student and its rollout context distribution. The benefit of modifying this moving target depends on how target--student probability mismatches translate into updates. We propose \emph{Self-Referenced On-Policy Self-Distillation (SR-OPSD)}, which constructs a normalized geometric target from the self-teacher and a frozen initial policy, then minimizes the forward R\'enyi divergence from this target to the student. The interpolation coefficient controls the self-teacher's contribution, while the R\'enyi order controls the power weighting of target-to-student probability ratios in the gradient. For fixed contexts and target components, we establish a conditional variational characterization and derive the exact token-logit gradient, revealing how anchoring and projection jointly shape the effective update target. Experiments across scientific reasoning, tool use, mathematical reasoning, and code generation demonstrate strong performance across multiple model families and scales. Ablations further show that reference anchoring can improve or degrade performance depending on the projection objective, supporting the joint design of target construction and projection geometry.

Neural networks can acquire new capabilities while damaging existing ones, but what determines whether new learning persists remains unclear. We identify functional compatibility, the extent to which incoming learning can coexist with behaviour that must be preserved, as an experimentally manipulable causal determinant of persistence. From identical neural states, we vary compatibility while matching unrestricted learning opportunity and imposing a common retention requirement. Persistent learning increases with compatibility across independent directions, convolutional and transformer architectures, vision and text, and a ten-seed replication. Learning rules and retention constraints determine how much compatible opportunity is retained, whereas nonlinear geometry limits the matched intervention at larger update norms. Functional compatibility therefore reframes stability-plasticity from preventing forgetting to determining which new learning can coexist with existing function and persist.

Reinforcement Learning with Verifiable Rewards (RLVR) and on-policy distillation (OPD) have become two widely adopted paradigms for post-training large language models. However, RLVR suffers from sparse task-level feedback, while OPD provides dense token-level guidance but ignores trajectory correctness, limiting its performance to that of the teacher. Combining them is a promising direction: OPD supplies dense supervisory signals, while RLVR provides task-level correctness. Nevertheless, existing integrations often rely on weighted combination or heuristic switching, introducing extra hyperparameters and trade-offs. We propose On-policy Distillation with Verifiable Reward (OPDVR), a simple yet effective method that seamlessly combines OPD and RLVR without adding any hyperparameters. We first reformulate the implicit reward of sampled-token OPD based on trajectory correctness, then apply a ReLU gating mechanism to ensure that correct trajectories receive non-negative rewards and incorrect ones receive non-positive rewards---thereby aligning the distillation signal with task success while preserving the teacher's distributional guidance. Furthermore, our modification transforms sampled-token OPD into a proper RLVR method, making it readily combinable with any policy gradient algorithm, such as GRPO. Experiments on six reasoning benchmarks show that OPDVR consistently outperforms standard OPD. Our code is available at https://github.com/LeapLabTHU/OPDVR.

The functional annotation of genes in non-model organisms remains a significant challenge in computational biology, with 20-70% of sequenced genes lacking characterized functions. Traditional homology-based methods are often costly and strongly dependent on high sequence similarity. This study presents Homo-RAG, a framework for large language model-based gene function prediction that integrates homology-guided multi-hop retrieval with evidence-aware ranking. The framework exploits biological relationships between zebrafish and human orthologs to guide evidence acquisition from ZFIN, UniProt, and PubMed through hybrid dense and lexical retrieval. An Evidence Confidence Score (ECS) integrates semantic relevance, entity matching, orthology information, source reliability, and literature association signals to refine the ranking of retrieved evidence. Extensive evaluation across 150 queries and 7,200 retrieved documents shows that evidence weighting parameter of lambda=0.50 improves NDCG@10 to 0.9879 and MRR to 0.99, while retrieving relevant evidence for 99.33% of queries. Furthermore, 80% of the retrieved documents are query-exclusive, indicating that evidence quality complements rather than replaces retrieval relevance. These findings establish Homo-RAG as a practical and robust framework for reliable, evidence-grounded gene function prediction in understudied organisms. The study addresses important limitations of conventional annotation pipelines while identifying opportunities for future improvements in evidence features and attribution mechanisms.

Auto-research agents now run machine-learning development unattended for hours, revising data pipelines, models, and validation from their own feedback, yet on most competitions they still finish below strong human competitors. Outcome-based benchmarks record this gap but not its cause, because they grade the final submission and discard the development process behind it. We introduce TraceML, which pairs human and agent work on the same competitions under one version-level schema: 4{,}465 human Kaggle trajectories across 134 competitions, seven of which are also worked by two agent scaffolds, giving 430 paired human and 207 agent trajectories. Every code version carries its score, its timestamp, and labels for the action taken, its intent, the edit size, and the score effect. Read this way, the gap becomes concrete. Experts alternate data work, validation, model changes, and ensembling, and return to approaches they had set aside. Each agent scaffold instead collapses into a narrow loop: Codex spends its steps re-weighting ensembles and tuning submissions, MLEvolve mutates its model in place, and neither pivots at the human rate nor reopens abandoned work. A short planning prompt distilled from human practice moves the behaviors it names toward the human profile and lifts scores, but the effort profile stays agent-shaped: instruction closes only the part of the gap that reduces to instructions. We release the corpus, the schema, and the labeling models, together with an open-source toolkit that turns a run from any command-line agent into a TraceML trajectory and reads it against the human cohorts.

Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in the long term. However, this claim has not been formalized so far. In parallel, performative learning provides a framework for studying learning systems whose deployment affects the data they later observe. In this work, we bring these two perspectives together and introduce performative privacy, where data leakage reduces future participation. We study a simple model where agents repeatedly contribute data for mean estimation but may leave the system when their data is leaked. Privacy is implemented through differentially private mechanisms, creating a trade-off between estimation noise and future participation. We show, through a theoretical study of the dynamics and numerical experiments, that a finite privacy budget can outperform non-private estimation in the long term when the feedback loop between leakage and participation is sufficiently strong. This provides first evidence that differential privacy can be optimal not only as a protection mechanism, but also from the perspective of long-term utility.

Modern artificial-intelligence coding agents can be equipped with tools for checking their own work e.g. a linter, a boot probe, a shell, a screenshot tool. We call this set the agent's verification surface. This study asks whether increasing only that surface, with everything else held fixed, produces a matching growth in the quality of the software the agent ships. We built a minimal coding agent whose tool list is the single controlled variable and used it to implement 1,116 web applications across six models and eight tool configurations. A condition-blind human graded every application against a frozen rubric, and automatic probes stress-tested the API-observable behaviors. Verification's cheapest benefit arrives first, which is to make sure that the application comes up. Without any tools, about one build in seven fails to launch at all and a single boot probe removes nearly all of these failures at roughly 35 percent of a full shell's token cost, while the full shell multiplies the no-tools cost by 2.35. Screenshots help most where mistakes are visible (e.g. element placement, interaction), though even there the gain over a shell is modest and does not survive correction for multiple statistical comparisons. In cases where failures can only be measured rather than seen, such as keeping scrolling smooth over a 100,000-row list, screenshots add nothing. A verification tool improves the output artifact only where its reach covers the way the application actually fails.

Specialized machine learning architectures encode structural assumptions (equivariance, permutation invariance, relational structure) that language-based foundation models lack by design. This review asks whether such assumptions can instead be acquired through language, drawing on a corpus of 186 papers published between 2016 and 2026 across nine modalities: tabular data, graphs, time series, vision, chemistry, code, knowledge graphs, point clouds, and protein structure. Methods are organized into eight representational regimes, ranging from language-only prompting to fully specialized architectures, and are assessed not only on predictive accuracy but on whether structural information is preserved by the representation and computed by the model. Language-mediated models prove competitive in extreme few-shot prediction, discretized symbolic tasks, and textually annotated knowledge graphs. No evidence of general replacement survives once structure itself is evaluated: where direct tests exist, apparent parity is traceable to information asymmetry, benchmark contamination, or computation performed outside the language model. Across independent research communities, missing structural inductive biases are reintroduced through graph modules, structure-aware tokens, or specialized attention, indicating that specialization is relocated rather than eliminated. The review closes by specifying a falsifiable controlled experiment that would measure the remaining gap directly. The derived data supporting the findings of this review are openly available in the repository at https://github.com/kiyan-rezaee/language-vs-structure.

Closed-loop AI scientists can generate candidate designs at low marginal computational cost, whereas reliable feedback may require wet-lab synthesis, characterization, or high-fidelity computation. Addressing this imbalance through custom laboratory automation remains infrastructure-intensive and costly, while replacing new experiments with a fixed surrogate leaves persistent model errors that can be amplified by optimization. We propose \emph{online surrogate repair} (OSR), a closed-loop algorithm that uses sparse high-fidelity evaluations to update the surrogate throughout a longer agent search conducted primarily with inexpensive surrogate feedback. An acquisition rule selects which designs from the agent's accumulated proposals receive high-fidelity evaluation, and the resulting labels update the surrogate used in subsequent episodes. Across controlled synthetic environments, we demonstrate that improving global surrogate fit does not necessarily reduce maximum regret, whereas Q90-UCB and expected improvement (EI) substantially reduce regret by directing evaluations toward regions that determine the optimizer's decisions. On MADE, controls receiving high-fidelity feedback after every episode require $6.36$--$7.23\times$ more oracle queries to match Online EI under two LLM orchestrators and $10.27\times$ more under the non-LLM Chemeleon+MLIP workflow. Online surrogate repair introduces a novel third feedback regime between fixed-surrogate operation and high-fidelity feedback after every episode, separating the frequency of high-fidelity evaluation from the duration of the agent's search.

Linear attention enables efficient long-context inference by compressing token history into a fixed-size recurrent memory. This compression makes each update a trade-off between incorporating new information and preserving useful associations. Models such as DeltaNet, Gated DeltaNet, and KDA predict write strength from the current token representation, without explicitly tracking uncertainty in the stored memory. Yet this uncertainty matters: a new observation should have greater influence when the existing association is uncertain and less when it is already well supported. We introduce Kalman Delta Networks (KDNs), a family of linear-attention models that explicitly track memory uncertainty to guide each update. By formulating associative memory as a linear-Gaussian state-space model, KDNs propagate both the memory estimate and its uncertainty, using the Kalman gain to balance accumulated evidence against the reliability of new observations. This formulation also recovers standard delta-rule updates by replacing tracked covariance with a token-predicted isotropic surrogate. To support hardware-efficient training and inference, we derive Diagonal KDN and Isotropic KDN, which retain one uncertainty value per key channel and per head, respectively. Their uncertainty updates admit associative scans with logarithmic parallel depth, requiring only $O(d_k)$ and $O(1)$ auxiliary state per head. Across controlled pretraining at 750M and 1.3B parameters, both variants consistently improve perplexity and mean downstream accuracy over the evaluated state-of-the-art linear-attention baselines.

Test-time reinforcement learning (TTRL) enables models to improve their reasoning without relying on labeled training data, but existing approaches typically optimize a large fraction of the model parameters. This raises a natural question: can effective test-time adaptation emerge when both the reward signal and the optimization space are severely restricted? We answer this question with label-free bias-only TTRL, which uses majority-vote pseudolabels as rewards and optimizes only ~100K bias parameters while keeping the pretrained backbone frozen. On MATH-500, our approach reaches 76.67% accuracy with Qwen2.5-7B, slightly exceeding our own labeled bias-steering reproduction while optimizing 76,000x fewer parameters than full-parameter TTRL. The same training procedure improves performance across vision-language and audio reasoning tasks, including MathVista, AI2D, LogicVista, and MMAU. We further show that the learned steering vectors transfer to 4,500 held-out MATH problems, indicating that the adaptation is not limited to the problems used during test-time optimization. Finally, we analyze why this highly restricted adaptation can work, showing that majority-vote reliability improves with rollout consensus and that bias subspaces with greater accessible gradient energy exhibit stronger downstream trainability. These results demonstrate that substantial test-time adaptation can emerge from optimizing a tiny bias-only subspace using entirely label-free rewards.

Neurological prognostication after cardiac arrest commonly relies on electroencephalography (EEG). However, EEG demands high clinical resources. Bedside electrocardiography (ECG) is standard and low-cost. Yet, its value for predicting neurological outcomes remains underexplored. In this study, we propose NeuroECG, an ECGFounder-based deep representation framework for EEG-free auxiliary prognostication. NeuroECG adapts a pretrained ECG foundation model via task-specific fine-tuning. We implement a gradual unfreezing strategy on single-channel bedside monitoring ECG. Multiple ECG segments per patient are encoded into segment-level deep features. These embeddings are aggregated via quantile pooling (q = 0.24) and compressed using principal component analysis (PCA). Experiments on 412 ECG-available patients from the multicenter I-CARE database show that the adapted ECGFounder backbone achieves the best performance among ECG-only backbone baselines, with a test AUROC of 0.7333. We further combine the learned deep ECG representation with static clinical covariates. The proposed NeuroECG model achieves a test AUROC of 0.8077 and an AUPRC of 0.8970. These results support deep bedside ECG representations as a useful source of auxiliary prognostic information. Their integration with static clinical covariates improves prediction in an EEG-free setting. The source code is available at https://github.com/goddream66/NeuroECG

Complex robotic manipulation tasks frequently require a long-term memory of past events and actions. As conditioning on full histories renders policies prone to spurious correlations and degrades performance, many approaches to policy memory involve compressing historical information through expensive VLM queries in-the-loop to process only task-salient information. In this paper, we propose an alternative approach in which computationally intensive VLM queries are made during train-time to learn a lightweight latent memory that can be efficiently queried at deployment time. Our representation, which we call the \textbf{workspace token}, is trained by (1) using a VLM to identify current and historical information necessary for completing a task, then (2) distilling these into the workspace token using a set-reconstruction decoder loss. In both simulation and hardware, we show that the workspace token can be used as a drop-in replacement for observations during deployment, enabling policies to solve memory-intensive tasks without the need for VLM reasoning in-the-loop, in effect serving as a \textbf{latent harness} for distilling a stronger reasoning models ability to solve long-horizon tasks to a reactive robotic policy. We further demonstrate that the workspace tokens are not only more lightweight, but also lead to better policy performance compared to conditioning policies on explicit modalities like curated past image frames, motivating a ``latent'' approach to history curation and reasoning model harnesses more broadly.

Autonomous parking in nonconvex and narrow environments remains challenging. Although optimal-control methods can explicitly enforce vehicle dynamics and collision constraints, nonconvexity compromises solver robustness and can cause failures. Large language models (LLMs) exhibit strong semantic reasoning capabilities, but directly generating dense trajectories makes it difficult to guarantee physical feasibility. We introduce SE-LLM-OCP, a unified framework in which LLMs make high-level discrete maneuver decisions, while an optimal-control module enforces low-level vehicle dynamics and collision constraints. Online, the LLM proposes sparse maneuver plans, decomposing the parking task into a sequence of short-horizon trajectory-optimization problems. A low-level solver then sequentially solves optimal-control problems. If the solver fails, the LLM aggregates failure evidence from the solver and validation stages to guide replanning. Offline, SE-LLM-OCP automatically evolves a structured decision-making knowledge base from scratch, driven by accumulated online failures. We validate our proposed framework in simulation on a car-like vehicle model and on a differential-drive robot. Our experimental results show that SE-LLM-OCP enables safer autonomous parking in narrow scenarios and demonstrates transfer of the same maneuver representation to a different kinematic platform.

Industrial control system attacks are usually documented in terms of the plant where they occurred: its sensors, actuators, process stages, and control logic. Yet many attacks express a more general physical pattern--such as suppressing flow, corrupting chemical dosing, or driving a vessel toward overflow--that may also matter in a different plant. The challenge is deciding when such a threat remains meaningful on a new system rather than relying on similar component names or broad semantic labels. We present XPhysICS, a methodology for grounding documented cyber-physical threats onto a specific target system. XPhysICS converts source evidence into a provenance-linked description of what is manipulated, what physical consequence is expected, and what observations the evidence calls for. Once this analyst-guided abstraction, its vocabulary and schema version, and a target contract are fixed, XPhysICS applies deterministic grounding checks. An accepted result can be represented as a validation slice that records the mapped roles, signals, dependencies, and context intended to support later evaluation. We study 83 threat abstractions across continuous-process and manufacturing sources using separate evaluation denominators. The continuous-process study evaluates 78 abstractions against target contracts spanning water treatment, water distribution, hydropower, and chemical processes. Selected cases are exercised through controlled perturbations of simulator-role signals. We also test compatibility with several analysis styles, including the released upstream GeCo implementation, and conduct a three-objective, one-target realizability study using a paper-derived search reproduction. Across these evaluated settings, the results support treating explicit target checks and traceable evidence as separate from semantic similarity alone.

Vision-and-language navigation increasingly relies on general-purpose semantic planners, yet translating correct high-level intent into reliable physical execution remains difficult in spatially constrained transitions. Reaching a staircase, doorway, or narrow passage does not ensure traversal; the agent must identify an executable affordance pose and recover from accumulated action errors. We propose PACE (Preference-refined Affordance-Conditioned Execution), a supervised local execution module that augments frozen zero-shot semantic planners for reliable cross-floor navigation. PACE grounds transition-related semantics into a long-horizon, agent-centric traversable affordance pose and conditions short-horizon action generation on this spatial target, thereby aligning semantic goals with physical execution. We further post-train PACE through failure-aware preference refinement using rollout-derived pairs that contrast normal or recovery behaviors with deviation-amplifying behaviors, thereby improving closed-loop correction. We integrate PACE into six open-source zero-shot VLN navigators and demonstrate consistent improvements on the cross-floor subsets of R2R-CE and RxR-CE, increasing the average success rate from 16.35% to 27.65% and from 4.76% to 12.06%, respectively. Real-world experiments further demonstrate PACE's applicability in unseen environments, highlighting the potential of traversable affordances to bridge semantic intent and reliable embodied behavior.

Robotic manipulation is inherently history-dependent, yet most pretrained robotic policies condition on only the current observation or a short temporal window. Equipping such policies with long-term memory remains challenging: existing approaches either feed the backbone multi-frame observation windows, which substantially increase inference cost, or rely on pre-defined semantic features, which limit task generality and may also require the retraining of the backbone to adapt to the memory. We introduce DRAM (Delta-rule Recurrent Associative Memory), a plug-and-play memory module that can be attached to a wide range of pretrained robotic policies, endowing them with long-horizon memory without architectural modification or backbone retraining, requiring only task-specific post-training of the memory module and action expert. DRAM maintains a fixed-size associative memory using gated delta-rule linear attention, with a modified update that incorporates all tokens within each frame in parallel. An architecture-agnostic readout integrates historical context into action prediction across different policy architectures. Experiments show that DRAM consistently improves frozen pretrained policies over short-context baselines and alternative compact memory designs, validating its effectiveness as a fixed-size, post-hoc memory module trained with the backbone frozen.

Traditional scientific computing requires researchers to translate computational intent into environment configuration, resource requests, and executable jobs, then diagnose failures from scheduler state and application logs. We present ASCEND (Autonomous Scientific Computing Engine and Novel Discovery), an AI-powered agent interface that runs the agent on the researcher's own laptop, reaching Slurm-managed clusters and a GPU workstation over a multiplexed authenticated connection, with site-specific execution policies checked by locally executed tools; the language model is hosted remotely and holds no credentials. No facility-scale service is required: an account on each resource is sufficient, and the public installer lets users link additional Slurm clusters or workstations of their own. We report four recorded cases: (1) the agent closed a failure-recovery loop on a planted tensor-device fault, submitting, diagnosing, repairing and resubmitting with job-level artifacts preserved; (2) it reproduced the published evaluation of a weather-forecasting model from the author's released forecasts, agreeing with the published curves to 2.1% (z500) and 2.4% (t850) while identifying a unit discrepancy in the paper's prose and an initialization-field discrepancy in its released data; (3) it parallelized a released 12,693-line geophysical solver under a bit-for-bit identity requirement, reducing wall-clock runtime from about twelve hours to about two; (4) that requirement exposed two instances of undefined behaviour in the published solver, both repaired and reported upstream. Separately, a pre-specified evaluation of the policy layer found the deployed validator rejected 29 of 30 constructed violations and held the remaining one for approval, while denying 3 of 14 legitimate requests. Autonomy was exercised under author supervision; an end-to-end recovery benchmark remains outstanding.

World Action Models (WAMs) are becoming increasingly important and useful for embodied intelligence, as they enable robots to anticipate the consequences of candidate actions before interacting with the physical environment. However, underwater robots are usually subject to passive dynamics, such as inertia, buoyancy, hydrodynamic drag, and persistent drift, which can continue to affect the vehicle even after an action is completed. Existing WAMs, which primarily predict action-conditioned visual observations, are not explicitly designed to capture such passive motion dynamics. In this paper, we present AquaWAM, the first World Action Model designed for underwater embodied agents. Instead of predicting future images, AquaWAM models both action-conditioned and passive physical dynamics, including the thruster dead band, the inertial glide that outlasts each command, and ambient currents. Specifically, it senses through the DVL, IMU, pressure sensor and joint encoders, while cameras supply only semantics for understanding goals and target pose. By modeling compact navigation states rather than high-dimensional visual observations, AquaWAM substantially reduces the model size and computational cost compared with conventional WAMs. Experimentally, AquaWAM achieves a 72.6% task success rate across 20 underwater tasks on the USIM benchmark, outperforming existing methods while making action decisions 2.7x faster than U0 on an NVIDIA Jetson AGX Orin. Our model also remains effective when some onboard sensor measurements are unavailable. For example, without DVL velocity measurements, our method still achieves a 61.6% success rate, compared with 39.4% for U0.

Model-based judges support agent security by detecting prompt injections, assessing interaction risks, and screening harmful requests. System One models select from predefined answers and report probabilities that software can use to allow, block, or review inputs, but the reliability of these automated decisions remains unclear. We evaluate Jev, Laya, Decider, and Bespoke Nimble against specialized classifiers and language-model judges, examining decision accuracy, probability calibration, and selective automation. We draw the following conclusions. (1) Strong overall performance and favorable aggregate calibration can hide failures concentrated in particular attack groups, including attacks classified as safe with high confidence. (2) The evaluated adapted configurations do not consistently improve classification over their base models across tasks. (3) Under the strictest evaluated error limits, the policies allow few inputs automatically, and separate allow and block thresholds increase automation mainly through more blocks. Passing confirmation does not ensure that these limits hold on test. (4) Judges can detect attacks missed by another model, but may also falsely flag more benign inputs and share the other model's high-confidence errors. These findings support evaluating model accuracy, probability calibration, and the resulting allow/block/review decisions together.

Dexterous manipulation requires tactile feedback. However, robot tactile demonstrations are difficult to scale,because dexterous-hand teleoperation provides limited tactile feedback to the operator. In contrast, human demonstrations offer a substantially more scalable source of diverse tactile interactions. Motivated by a simple premise: hands can change, but the underlying physics of interaction does not. We leverage human tactile data to improve dexterous manipulation policies. Specifically, we first build a tactile motion-capture system that synchronously records images, tactile signals, and hand motions. Using this system, we construct the UVTA dataset spanning five contact-rich tasks, with 1,000 human demonstrations covering diverse interaction patterns and 150 robot demonstrations per task. To transfer the underlying physics of human interaction to robot control, we propose a Unified Visual-Tactile-Action Model that maps both embodiments into aligned tactile and action representations and jointly predicts future action and tactile trajectories. The joint objective enables human demonstrations to supervise contact-aware representation learning, while only robot actions are executed during deployment. In real-robot evaluations across five tasks, our method achieves an average success rate of 70%, outperforming the strongest visual-tactile baseline, which achieves 29%, and an architecture ablation, which achieves 42%. Performance improves consistently with additional human demonstrations and exhibits no saturation at 1,000 demonstrations per task, validating the effectiveness of scalable human tactile data for dexterous manipulation. Project page is available at https://uni-vta.github.io/.

LLM agents operate in persistent collaborative environments involving multiple users, communities, memories, files, and tools. Community boundaries may remain fixed or evolve with changes in membership, roles, composition, and relationships. Agents must complete legitimate tasks and prevent unauthorized disclosure of protected information. Existing evaluations do not fully examine these risks in agent systems. We introduce \textbf{CoSec}, an executable benchmark for evaluating privacy and authorization enforcement in LLM agent systems operating within and across communities. CoSec contains 208 canonical scenarios spanning fixed and evolving boundaries, protected information belonging to the agent owner or other participants, and attacks through dialogue, environmental content, persistent memory, and composed workflows. CoSec executes complete agent systems with persistent sessions, memory, files and tools. It verifies information flows against the active authorization state using execution traces and artifacts. Across harness and model configurations, agents frequently complete benign tasks but violate privacy and authorization boundaries. Privacy behavior varies across harnesses, attack surfaces, and community states, revealing how memory, files, tools, and workflows can carry protected information beyond its authorized scope. These findings show that task utility does not imply privacy or authorization compliance and that authorization in community settings remains an unresolved security challenge for persistent LLM agents.

Many Model Context Protocol (MCP) servers wrap web APIs built for human developers, and their error messages tell the reader to run a command, edit a configuration, open a web page or wait. Many agents that read them can only call the server's tools. In 150 widely used MCP servers, 949 of 3,001 error messages tell the caller what to do next, and half of these steps depend on something the server cannot see about the caller. On credential errors, 62 of 67 steps ask for a terminal command, a configuration change or a web page; on rate limits, 20 of 30 say to wait and retry without naming the call to repeat. We tested five OpenAI models that act only through the tools of Berkeley Function Calling Leaderboard tasks, and the agents did what the step said. On expired credentials, a terminal command in the step left 45% of tasks recovered, and the loss it caused grew from 18 points for GPT-5.5 to 69 for GPT-6 Astra. On a rate limit, GitHub's "Wait before retrying." left 6%. We tested two remedies. For MCP developers, naming a server tool in the step raised recovery on expired credentials to 84%, with the login tool in place of the command, and on a rate limit to 88%, with the call to repeat in place of the bare wait. For agent developers, deleting the step with a one-sentence prompt before the model reads it raised recovery on expired credentials to 82%.

The real-world performance of current vision-language-action models is fundamentally constrained by the limited coverage of expert demonstrations and their insufficient understanding of physical interactions. A common remedy is to collect additional real-world demonstrations of newly encountered failures. However, this process is costly, inefficient, potentially unsafe, and difficult to scale. To address this challenge, we propose Failure for Rising (F4R), a failure-driven real-to-sim-to-real closed-loop learning framework that converts real-world failures into targeted policy improvement. F4R first uses an agent to automatically identify and diagnose failures from rollouts. It reconstructs each failure as an interactive, object-centric table-top environment that preserves the task-relevant spatial and physical conditions. The policy is then refined through failure-conditioned sim-real co-training followed by targeted reinforcement learning in the reconstructed environments. The improved policy is subsequently redeployed, while newly observed failures are continuously fed back into the next reconstruction and learning cycle. Real-world evaluations on four manipulation tasks show that F4R achieves 93.75% In-Distribution and 90.0% Out-of-Distribution (OOD) success, outperforming the budget-matched Targeted BC baseline by 18.75 percentage points under OOD conditions without collecting additional real-world corrective demonstrations.

Structured phenotype extraction from radiology reports supports cohort construction, quality auditing, and clinical analytics, but practical deployment requires local inference and auditable predictions, while expert annotations remain scarce. Conventional labelers provide structured findings and assertion states but no supporting evidence, while API-hosted large language models may be unsuitable when clinical text cannot leave institutional infrastructure. We present CHESTPHENOT, a compact 0.5-3B language model that jointly extracts finding labels, three-class status (present/absent/uncertain), and verbatim supporting evidence spans. CHESTPHENOT is trained using hybrid CheXbert+72B silver supervision followed by supervised fine-tuning and lightweight GRPO refinement. Across three human-annotated gold sets spanning in-distribution, cross-taxonomy, and cross-institution evaluation, the 3B model remains below its CheXbert silver teacher in distribution but is competitive under distribution shift, significantly surpassing CheXbert on cross-institution detection (+2.0 F1). Task-specific training also enables the 3B model to match or exceed substantially larger prompted models on most detection and status comparisons. For evidence-grounded extraction, over 99% of final evidence spans are locatable in the source report, and the 3B model achieves 47.5 auditable-F1, outperforming Qwen2.5-7B one-shot prompting by 7.6 points and approaching Qwen2.5-72B. These results demonstrate that locally deployable models can provide competitive and directly auditable radiology-report extraction without relying on external inference APIs. Code and the full extraction/judge prompts will be made available at https://github.com/yukkai/ChestPheNoT.

Scientific literature contains latent knowledge about materials behavior, but much of this knowledge is expressed through words, contexts, and recurring associations rather than explicit design principles. This raises a central question: how can large-scale scientific corpora be used for practical problems in materials discovery? Here, we present a literature-guided descriptor-based filtering framework for reducing composition search spaces. For a given performance metric, the framework selects two descriptors from a filtered vocabulary in a literature-trained word embedding model and uses the selected descriptors to construct a Pareto-based filter for candidate compositions. Across the evaluated performance metrics and composition search spaces, the framework filters out an average of 74.27\% of the candidate compositions, with an average best-value error of 1.93\% relative to experimental measurements. Compared with expert-chosen and random descriptors, our performance metric-dependent descriptors provide a more controlled balance between retained fraction and best-value error. These results show that literature-derived embeddings can support intuitive and reproducible filters for narrowing candidate composition spaces while preserving high-performing compositions.

Sentiment analysis is a common tool for converting unstructured text into quantitative signals in finance and risk management. Yet most applications reduce the output to a discrete polarity label or a single predictive feature, overlooking the distributional structure of sentiment intensity in consumer complaint narratives. In this paper we treat negative sentiment in consumer complaints as a bounded continuous variable and study its full distribution rather than a single label. We score each narrative with a transformer classifier, model the scores with Beta distributions, and compare the fitted distributions of meritorious and non-meritorious complaints through the Kullback Leibler divergence and the squared Hellinger distance. The fitted distributions are then linked with dollar amounts and company response outcomes to construct anomaly diagnostics that flag complaints whose textual severity is inconsistent with the recorded relief. We find that the two groups have strongly overlapping distributions, so negative sentiment intensity is not a sharp classifier of outcomes on its own; combined with monetary and categorical attributes, it isolates unusually severe complaints for operational risk monitoring. Treating sentiment analysis as continuous distributional measurement, this study links sentiment extraction, bounded response modeling, and anomaly detection in a unified framework for consumer complaint assessment.

Retrieval-augmented generation (RAG) pipelines are typically assembled from independently-chosen components -- a document parser, a chunking strategy, and an embedding model -- yet these choices are rarely evaluated jointly, and evaluations that do combine them are usually run on a single document or corpus. We present a controlled factorial study of 3 parsers, 3 chunking strategies, and 5 dense embedding models, together with a sparse BM25 baseline, evaluated against 800 question instances, each with one or more required evidence strings, with evidence strings automatically validated against source text and a 10% random sample manually reviewed, across four structurally distinct Indian central-government regulatory documents. We fit linear mixed-effects models with document-query-level random intercepts to the resulting 72,000-row result set, run Holm-corrected paired comparisons between matched dense and sparse configurations, and report clustered bootstrap confidence intervals for all 54 unique retriever configurations. We find that no single retriever family dominates across documents; parser and chunker choice interact significantly; MPNet-base is a consistent underperformer with a severe failure mode on table-derived questions; and the corpus exhibits a near-saturated evidence-preservation ceiling above 98%, indicating that retrieval differences are driven primarily by ranking quality rather than information loss during ingestion. We additionally report embedding-dimension and chunk-size/overlap ablations and an efficiency/quality Pareto analysis. We release our full evaluation harness, corpus manifest, and 800-question benchmark.

Ontology-driven knowledge graph construction from industrial text remains challenging due to the domain specificity of documents, the scarcity of annotated resources, and the complexity of ontology engineering workflows. This paper presents and investigates the applicability of an ontology learning pipeline that combines compact open-source Large Language Models (LLMs), reusable prompting strategies, and open knowledge bases to support the extraction, structuring, enrichment, and evaluation of knowledge from textual corpora. The approach is tested and evaluated on a private French corpus of power-grid incident reports, using locally deployable open-source LLMs ranging from 7B to 32B parameters. Starting from unstructured reports, the approach extracts entities and relations, generates RDF triples, constructs related OWL ontology, enriches it using external knowledge sources, assesses the quality of the ontology, and subsequently constructs a populated knowledge graph grounded in the resulting ontology schema. Experiments on 80 manually annotated private reports show that schema-guided prompting significantly improves extraction quality, while quantized models provide an effective trade-off between performance and computational cost. These results demonstrate the feasibility of transforming domain-specific industrial text into ontology-based knowledge graphs using locally deployed open-source LLMs, while supporting the generalization of the extraction process through reusable prompting strategies.

Recent work suggests that obscure language registers can weaken large language model refusal behavior, especially when paired with black-box prompt optimization. It remains unclear whether failures stem from non-standard surface form, culturally grounded framing, or optimization over an expressive prompt-strategy space. We study Chinese registers by extending a classical-Chinese red-teaming framework to Shanghainese and Cantonese and running a 36-cell ablation across surface forms, strategy-bank variants, and two target models. The main finding is corrective: dialectal surface form is neither necessary nor sufficient for high attack success. Non-optimized English, Mandarin, and naive dialect translations remain below 8\% attack success rate, whereas all conditions that retain an optimizer-controlled strategy bank reach 98--100\%. A culture-neutral generic strategy bank reaches the same ceiling at near-single-query cost, further indicating that strategy-bank expressiveness, rather than dialectal or cultural content, accounts for most of the observed effect. Dialect choice still affects query efficiency and response severity, and qualitative coding identifies recurring high-level mechanisms such as semantic glossing and procedural scaffolding. We qualify the absolute ceiling-level attack success rate by showing sensitivity to judge calibration and noting that the design does not fully separate one-shot strategy construction from iterative search.

Digital pens are widely used to capture handwriting on digital devices, enabling precise trace recording and enhancing human-computer interaction. However, most are bundled with tablets and lack cross-brand compatibility. Recent digital pens equipped with kinematic sensors have emerged, designed for use on any surface. This especially opens significant potential for supporting handwriting acquisition in classrooms. Handwriting reconstruction from such an IMU-equipped pen poses a challenge due to the significant variability in sensor signals between adults and children. Even when producing visually similar traces, variations in writing dynamics, motor control, pen holding, and user confidence introduce substantial discrepancies in the captured signals. Additionally, the high variability in children's handwriting requires collecting large amounts of data, which is not feasible to implement in a school environment at scale. Furthermore, models trained exclusively on adult data fail to generalize to children's handwriting, and conversely, models trained on children's data perform poorly on adult writers. This crosspopulation degradation highlights the need for a unified model that can be deployed directly on the pen, without any user-specific adaptation. To address this issue, we propose a cross-domain learning, using an original neural network architecture based on a Temporal Convolutional Network and multiple prediction heads. The model is designed to be robust across age groups by leveraging shared features while effectively handling variability induced by differences in graphomotor development. This approach aims to improve handwriting trace reconstruction from sensor data, where each domain benefits from additional data provided by the other domain.

Public speaking is an essential skill in academic and professional contexts, but it often causes anxiety. Although several AI-based speech coaching tools exist, they typically provide generic feedback and lack model speeches for learning. This study addresses these gaps by developing a system that incorporates speech context into evaluation criteria, enabling more tailored feedback. It also generates improved model speeches in both text and audio formats to support example-based learning. The system was evaluated in a pilot study with seven undergraduate students over one to three weeks. Participants used the system repeatedly and practiced the same speech at least three times. Results showed improvements in public speaking performance for all participants. Filler word usage decreased, and anxiety levels dropped by 15.3% to 34.4%. Participants found the context-aware feedback and revised speeches useful and confidence-building. These findings suggest that context-aware feedback and model speeches can enhance public speaking skills and reduce anxiety. Future studies could explore video-based analysis of physical behavior during speeches.

Retrieval-Augmented Generation (RAG) introduces a specific failure mode in discrete diffusion language models: when retrieved context contradicts parametric knowledge, the iterative denoising process becomes a visible battleground between competing knowledge sources. We identify temporal semantic divergence as an observable for detecting these conflicts and introduce the Trajectory Variance Score (TVS), a simple and interpretable measure of this divergence. TVS computes the mean pairwise cosine distance of answer embeddings across independent stochastic denoising trajectories, capturing the temporal tug of war between parametric and contextual attractors. Requiring as few as two parallel inference runs, TVS is computationally lightweight. Across four diverse datasets (Synthetic, SciQ, PopQA, and CounterFact), a simple Logistic Regression classifier using TVS achieves $70.10\%$ accuracy and $0.7647$ AUROC on LLaDA. On Dream 7B, increasing the number of trajectories from two to five improves accuracy from $63.91\%$ to $69.62\%$. More complex sequential models provide only marginal improvements over the linear classifier. Evaluation across LLaDA and Dream 7B demonstrates that conflict-induced trajectory dynamics and their key properties transfer across distinct diffusion architectures.

Parallel numerical libraries such as PETSc are widely used in science and engineering applications where wrong results can have costly consequences. Despite this, numerical libraries are rarely formally verified. One of the challenges is the need for an expert to hand-write a specification and manually apply a verification tool, often requiring the development of a harness or driver, all of which can contain additional bugs that lead to false positives or false negatives during verification. With recent advancements in large language models (LLMs), it is tempting to generate such drivers and reference models automatically, but one-shot generation based on a simple prompt is brittle and leads to additional unverified code that needs to be audited. In this paper, we present an approach to use LLMs in a limited setting to generate a small, human-certifiable ACSL contract from the function's documentation, combined with a deterministic toolchain that supports a restricted ACSL profile and generates a driver that uses the CIVL verifier to check the implementation against the contract and, when available, an existing reference model. We demonstrate the pipeline end-to-end on three PETSc functions: MatAXPY (reference model already exists), MatAYPX (no reference model, so the certified contract is the sole oracle), and the non-compressing mode of MatFilter (no reference model, with more complex, conditional behavior). With this pipeline, we were able to discover a bug in PETSc's MatAYPX function that was previously undiscovered and had been present in the code since 1997.

In verifiable domains such as math and coding, finding one correct solution among many attempts can matter more than the pass rate of each attempt. Post-training can concentrate large language model outputs around a few modes, while increasing sampling temperature has limited effectiveness. We introduce Don't Repeat Yourself Supervised Fine-Tuning (DRY-SFT), a post-training method that increases output diversity and coverage: the probability of at least one correct solution among many attempts. DRY-SFT has two stages. First, for each problem, sequentially generate K solutions, showing the model all prior attempts and asking for a different solution. Second, fine-tune on each attempt independently, removing prior attempts from the context. The process uses no reward, verifier, or correctness filter. On HumanEval+, MBPP+, and DS-1000, DRY-SFT raises pass@100 by 10.8, 12.5, and 12.4 percentage points, respectively, at a small cost to pass@1. Structural diversity, measured by abstract syntax tree edit distance among passing solutions, rises significantly on all three benchmarks. DRY-SFT also solves 244 of 600 problems that the base model did not solve in the same 200 attempts. Across nine open-weight models, lower structural diversity of the base model significantly predicts larger DRY-SFT gains, indicating that the method is especially effective on more mode-collapsed models.

Oshiwambo -- a cluster of mutually intelligible Bantu languages spoken by over a million people across northern Namibia and southern Angola, and the home language of roughly half of Namibian households -- has, to our knowledge, no published machine-translation evaluation benchmark. Major commercial services (Google Translate, DeepL, Microsoft Translator), open multilingual MT models (NLLB-200, MADLAD-400), and the open Masakhane checkpoint collection all lack coverage of either standardised dialect, Oshindonga or Oshikwanyama. We announce Ongiini-Eval-OW, a planned 600-item English-Oshindonga and English-Oshikwanyama benchmark with native-speaker references from two independent translators, a 30-item inter-translator agreement set, an 11-tag phenomenon-tagged stratification (at least 30 items per tag), a deterministic 30% blind split, and a reproducible scoring protocol over chrF++, BLEU, and COMET-22, supplemented by a 50-item human-evaluation round. We document the empirical coverage gap, the dataset composition, the launch-leaderboard model matrix across American, European, and Chinese frontier and open-weight systems, and the contribution pipeline. The dataset is targeted for first public release in Q4 2026; this v1.0 concept paper announces the design and the call for participation. Data and code will be released under CC-BY-4.0 and MIT respectively.