arXiv AI

arXiv:2610.30017v1 Announce Type: new Abstract: This study introduces GuideEvolve, a self-evolving agent framework incorporating expert guidance and three autonomous regulatory mechanisms, collectively termed the 3Ts: Think Fast and Slow, Throwback, and Termination by choice. The framework addresses critical limitations in extant self-evolving agent paradigms through dual innovations: (1) a task-specific expert-evaluation-driven evolutionary pipeline, and (2) the integration of 3Ts mechanisms to facilitate cognitive optimization, performance recovery via temporal rollback, and adaptive termination protocols. Empirical validation across the MGSM and MMLU benchmarks demonstrates that GuideEvolve attains performance metrics comparable to current state-of-the-art models (89.8% accuracy versus G\"odel Agent's 90.6% on MGSM; 89.1% versus 87.9% on MMLU), while exhibiting notably reduced variance (25-59%) and enhanced robustness. Furthermore, the framework achieves a 100% evolution success rate, superior to the baseline's 82%, diminishes required optimization iterations by approximately 50% (from 30 iterations in baseline to 15.1 and 13.4 across tasks), and provides improved stability for production deployment.

arXiv:2610.29985v1 Announce Type: new Abstract: Benchmark performance is the dominant standard for comparing large language models (LLMs), typically summarized by a single aggregate score, such as accuracy, pass rate, or rubric score. This aggregate score provides limited information about a model's underlying capabilities, yet understanding these capabilities is crucial for model development and evaluation. To this end, we propose Tree-Induced Diagnostic Evaluation (TIDE), which recasts diagnostic evaluation as behavior-driven decision-tree induction to perform fine-grained capability diagnosis. TIDE recursively builds a hierarchical capability tree from benchmark samples and model scores: at each node, an LLM judge proposes candidate splits and performs sample-to-group classification, and behavior-based information gain selects among them. We apply TIDE to five models across four benchmarks to demonstrate its effectiveness, and find fine-grained capability contrasts invisible at the aggregate level: Claude Opus 4.6 achieves 100% on microeconomic market reasoning where GPT-5.4 scores 83.3%, but trails GPT-5.4 on contract law reasoning with 77.8% versus 88.9%. We hope TIDE can be extended to past and future benchmarks so that the rich behavioral signal in evaluation results is fully leveraged for understanding model capabilities.

arXiv:2610.29961v1 Announce Type: new Abstract: Given the enormous number of claims to be checked, verifiable legal AI requires scalable, cost-effective models that answer the practical needs of law. System One decision models are built on large language models but, instead of generating text, return a calibrated probability for each option. This suits the verification of legal authority, where the question is closed: a given holding, or none of them. We test whether such models abstain when the correct authority is missing. We extend CaseHOLD with an abstention condition in which the correct holding is replaced in 1,670 of 5,314 items, and compare four decision models with six generative LLMs. When no option is correct, generative LLMs still pick one in 38-86% of items; the most cautious decision model does so in 30%. On standard CaseHOLD, a decision model matches the best generative LLM (88.1% vs. 88.2%) at 1/55 of the cost. Accepting that LLM's answer only when a decision model agrees halves its wrong citations, and distilling a decision model's probabilities into Qwen3-8B, without labels, yields a well-calibrated local model. Legal AI should be judged by its wrong citations, not by accuracy alone.

arXiv:2610.29934v1 Announce Type: new Abstract: Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state. To address this challenge, we introduce ToM2Vec, which learns to form abstract theories-of-mind in a rich latent space via a fixed budget of continuous reasoning tokens. During training, supervision for these tokens comes from the partner's actual thoughts, which the ToM tokens are trained to abstract and match. At test-time, ToM tokens are generated by each agent and act as conditioners for that agent's thoughts and actions. Across several environments covering collaborative, competitive, and mixed-motive settings (Overcooked-AI, multiparty negotiation, Decrypto, Hanabi, and Collab Escape/Capture), we evaluate ToM2Vec's ability to support multi-agent interactions and collaboration, finding consistent improvements over the base model and a variety of competitive trained and prompted baselines. We show that latent reasoning is critical, with performance dropping when reasoning is expressed in text. Moreover, we show that ToM2Vec training generalizes to partner model families not seen at train-time, including a proprietary model (GPT-5.4). In our analysis, we provide evidence that the latent tokens encode a ToM, rather than task-specific information: ToM2Vec's tokens improve the agent's ability to directly answer questions about partner beliefs, desires, and next actions.

arXiv:2610.29822v1 Announce Type: new Abstract: As coding agents become more capable, developers increasingly use them to fix bugs through multi-turn interactions. Existing benchmarks do not adequately cover this scenario. Most present a complete issue report in a single turn without interaction. Their evaluation can produce both false positives and false negatives, accepting incorrect patches or rejecting correct ones. We introduce UserSWE for evaluating coding agents' ability to fix repository-level bugs through multi-turn interactions with dynamic users and execution-grounded verdicts at every turn. Its construction pipeline turns repository history and existing single-turn benchmarks into multi-turn tasks. At each turn, information pools constrain what the simulated user can reveal, while a persona chain sampled from a transition model calibrated on 20,859 development sessions governs how it speaks. Tests run after every turn, and a pair of agents from different model families writes and gates each verdict, using test results as primary evidence and model judgment only for aspects the tests cannot resolve. We evaluate eleven agents on 72 tasks from 42 repositories and find that agent rankings change with the interaction budget. Claude Fable 5.1 resolves the most tasks in one turn, GPT-6-Astra resolves the most by the eighth turn, and Claude Opus 5 leads at most intermediate budgets. These results indicate that single-turn performance is insufficient to characterize interactive coding ability.

arXiv:2610.29816v1 Announce Type: new Abstract: Incorporating AI systems into safety systems requires monitoring to avoid silent faults and ensure safe and reliable operation. Single monitors often do not provide actionable information for human supervisors, who act as the system's adaptive controller. We present COMET, a coordinated multi-layer runtime monitoring framework for human-supervised AI-enabled safety systems in which each monitor is anchored in a different fault mode. Monitor outputs are combined in a coordination layer and mapped by a communication layer to a traffic-light display. In this work, COMET consists of three monitoring layers: signal integrity, physics consistency, and the AI system's competence envelope. The framework is evaluated on an open dataset of compressed-air leaks, where an AI system detects leaks from acoustic emissions, with held-out conditions simulating covariate shift, semantic shift, and sensor faults, events expected during runtime but unknown in their exact appearance. All monitors are frozen after development and tested zero-shot, and the framework is compared against no-monitoring and single-monitor setups. The bare AI system fails silently and confidently under all three shift types, whereas COMET reduces the trusted-wrong-answer rate to zero on covariate-shifted no-leak data and on three of four sensor faults, at a nominal false-alarm rate of $\approx 1\%$. Single monitors come close in detection but cannot attribute the fault mode. COMET converts their anonymous detection flags into diagnoses. Residual silent misses under semantic shift define the boundary of passive monitoring. Runtime monitoring frameworks for AI-enabled safety systems must therefore combine quantitative fault detection and isolation with qualitative, human-oriented feedback.

arXiv:2610.29812v1 Announce Type: new Abstract: Literature-searching language agents are becoming a standard component of scientific assistants for the life sciences, yet there is little controlled evidence on which parts of their harness matter. We run a controlled ablation on biomedical claim verification (SciFact) with the live PubMed API as the only tool. Holding the backbone and output format fixed, we compare four harnesses: closed-book answering, single-shot retrieval using the claim as the query, a ReAct-style agent that writes its own PubMed queries, and an oracle given the gold abstracts, each with two efficient frontier backbones. Naive single-shot retrieval finds the gold paper for only 6% of claims and turns overconfident errors into abstentions without improving accuracy. Letting the model formulate its own queries triples gold recall and yields the best verdicts (62% accuracy vs. 49-59% for the other realistic harnesses), although these differences are not significant at our sample size and agents use fewer than half of their search budget. With gold abstracts, the same models reach 79-86% accuracy, a significant 17-24 point gap showing that retrieval, not reasoning, is the main bottleneck. Finally, a citation audit shows that closed-book models cite PubMed IDs that almost always exist (327/335) but are almost never relevant: a cross-family LLM judge rates only 3 of 327 existing citations as relevant to the claim, versus 67% for citations produced by the search agent. Existence checks are therefore insufficient safeguards for scientific agents. We release code, prompts, and all cached tool calls at https://github.com/saqlain2204/real-pmids-wrong-papers.

arXiv:2610.29680v2 Announce Type: new Abstract: Using an LLM to generate training data for another is among the most general ways to train one, and it remains common practice. Some frontier teams have closed that route for their own models: what comes back is an answer, or a short summary, without the detailed reasoning trace behind it. That is a little ironic: they were themselves built on distilling human knowledge, and distilling an open-source model is still entirely routine; the restriction applies to their own outputs alone. Recent work recovers the hidden reasoning by replaying encrypted blocks across models. It somehow works, but it breaks the providers' terms of service, so it is not something a product can be built on long-term. We built the alternative. We take the context and the teacher's answer as input, and generate the reasoning trace ourselves, in the style of the student's own model family. Training is two-stage, supervised and then adversarial: a generator learns to regenerate the family's own reasoning traces from their context and answer, and a critic learns how real a reasoning trace looks. Applied to a teacher's answer, the generator writes one in the family's style. An independent judge rates how likely traces are to be genuine on a scale from 0 to 1: ours average 0.68, real traces 0.66, and traces selected by the critic from four candidates 0.82. These are authenticity scores, not correctness rates. We test the idea on five benchmarks with everything else held fixed. On general tasks the five training corpora are about the same; on coding, the generated reasoning beats Fable 5's own thinking content. With the same 5,000 questions and the same teacher answers, HumanEval+ improves by 3.7 points and MBPP+ by 7.1 at the final checkpoint, and the MBPP+ lead holds at every checkpoint. We suspect that reasoning written in the student's family style is easier for it to learn from.

arXiv:2610.29646v1 Announce Type: new Abstract: Choosing among robot demonstration corpora is a ranking problem: an upstream diagnostic must preserve which candidates produce better downstream policies, not merely correlate with average performance. We study paired instruction dependence, the rate at which a demonstrator changes its first decision when only the goal changes. Across 136 independently seeded quadruped-navigation corpora, predictions developed on 56 corpora are evaluated prospectively on 80 held-out corpora. The audit is reliable and associated with downstream paired correctness, and lowers mean absolute prediction error when added to a correctness-aware baseline, but the primary proper-log-score increment is +0.059 nats (95% corpus-bootstrap CI [-0.027, 0.133]) and the prespecified top-20% comparison does not establish an incremental selection gain. An exploratory response decomposition suggests a structural source of this predictive-ranking mismatch: under the tested predictor specification, separating both-correct and both-wrong responses yields larger held-out gains than their unsigned switch statistic. Seed-to-seed agreement indicates that the held-out endpoint retains recoverable rank signal. These results show that counterfactual response structure can be reliable and predictive while providing limited incremental decision value in this setting when compressed into an unsigned statistic, motivating diagnostics that preserve outcome-relevant structure and are validated against the decisions they are intended to support.

arXiv:2610.29501v1 Announce Type: new Abstract: Parallel thinking has emerged as a promising paradigm for enhancing the reasoning capabilities of Large Language Models (LLMs) by exploring multiple reasoning paths simultaneously. However, we identify that parallel thinking via Reinforcement Learning (RL) typically suffers from the mode collapse of reasoning paths, where parallel thinking trajectories degenerate into similar ones. This homogeneity undermines the fundamental motivation of parallel thinking, that is, maintaining multiple hypotheses concurrently to prevent premature loss into a suboptimal solution. In this paper, we propose Diverse Parallel Thinking (DPT), a simple yet effective RL approach for parallel thinking designed to encourage path diversity. DPT rewards the model to develop distinct reasoning strategies by penalizing the prefix similarity of parallel thinking paths, thereby enhancing a more comprehensive exploration of the reasoning space. We conduct various experiments and analyses across multiple reasoning benchmarks to demonstrate the effectiveness of DPT. Results show that DPT successfully outperforms previous parallel thinking methods, achieving state-of-the-art performance over baselines. We validate path diversity as a key element to improve parallel thinking. Our code is available at https://github.com/JiaruQian/DPT.

arXiv:2610.29493v1 Announce Type: new Abstract: We determine the minimum number of directional commands needed to solve alevel of Order of the Sinking Star, modeled as a crystal-exchangepuzzle on an 11*11 board. A command swapsthe player with the nearest crystal in the selected row or column, regardlessof distance; only an empty ray permits an ordinary unit walk. Two gates admitthe player to the exit exactly while all sixteen crystals occupy their targetcells. We give an explicit 93-command solution and exclude every shortersolution by an exact, cost-bounded breadth-first search. The search representsa position by its seventeen occupied cells and one active cell. Its admissibleheuristic partitions command costs between a compressed core abstraction andresidual motion, strengthened by a necessary activation detour and a cornervisit condition. The completed computation retained 605,349,215 occupied setsand 1,863,925,136 states. We prove the reductions and pruning conditions,describe the exact storage scheme, and provide source code, execution logs,and a reproduction driver. The result is a computer-assisted proof for thespecified instance, with an explicit implementation trust boundary.

arXiv:2610.29491v1 Announce Type: new Abstract: A model's failures can reveal gaps in the behavioral criteria guiding its training. We introduce RubricOpt, a framework that lets a policy revise these criteria within preference optimization. The policy diagnoses its own rubric-free outputs, retains requirements already covered by the rubric, and edits clauses that leave recurring failures unaddressed. A shared semantic representation connects task demands, preference-data contrasts, and rubric clauses, while a deterministic gate checks evidence references and fixed semantic anchors. The rubric then guides Direct Preference Optimization (DPO) during training, with preference pairs, labels, and the objective held fixed, and is removed at evaluation. Our analysis relates the local effect on rubric-free success to semantic alignment and a residual beyond the measured factors. Experiments on capability and open-ended generation benchmarks show that rubric conditioning can improve outcomes over ordinary DPO, with gains and regressions varying across models and datasets. A matched continuation isolates the contribution of revision from additional training and finds mixed incremental effects for the tested edit, whose feedback overlaps the evaluation set. RubricOpt thus brings the behavioral specification itself into the policy's feedback loop.

arXiv:2610.29467v1 Announce Type: new Abstract: Long-horizon dialogue agents need to keep user memories stable enough for personalization while adapting when the user genuinely changes. Existing memory systems often append every observation or overwrite old entries by heuristic rules, which makes transient noise hard to distinguish from real belief drift. We introduce PreCoMem, a memory consolidation framework inspired by predictive coding that represents memory as confidence-weighted beliefs about the user's state. Its core signal, Effective Surprisal, combines semantic deviation and contradiction evidence, then downweights this signal when retrieval is diffuse and unreliable. According to effective surprisal, a two-threshold gate maps each turn to one of three deterministic updates: MAINTAIN reinforces supported beliefs, PROFILE stores ambiguous cues as low-confidence hypotheses, and CORRECT softly decays reliably contradicted facts. The PROFILE stage acts as an explicit waiting room, preventing premature commitment to noisy evidence. Experiments on LoCoMo, LongMemEval-S, and PersonaMem-v2 show consistent state-of-the-art accuracy; our new DiMoBench confirms that PreCoMem adapts to non-stationary belief drift without over-reacting to noise. The code is available at https://github.com/ZhongWishing/PreCoMem.

arXiv:2610.29435v1 Announce Type: new Abstract: World models are increasingly used for planning, where a planner ranks candidate actions by returns imagined by the model. However, these models are judged mainly by how well they render frames, even though a visually faithful rollout can still assign the wrong return. To identify where this mismatch arises, we introduce a transition-level diagnostic that replaces imagined transitions with real ones and measures the reduction in return error against a budget-matched random control. Our analysis yields two findings. (1) Event transitions, the few steps where a reward is earned or an episode ends, are rendered as well as other steps (median pixel-error ratio 0.85), yet restoring them removes a median of 70% of the return error. In contrast, restoring transitions with the largest MSE, SSIM or LPIPS error removes less than a random choice. Across the 44 Atari-100k model--collection pairs that pass our reliability checks, including three released families, event patches outperform random patches in 39 (sign test p=1.4 x 10^{-7}), with a median gain of +44 percentage points. (2) The remaining error is structured: a complementary patch separates errors at true events from those elsewhere, while an eight-step timing tolerance recovers few events missed by our deterministic models. This structure also guides repair: a rule recorded before four repair outcomes were known predicts when a diffusion model repairs these failures. Event patches improve plan ranking, and the same concentration appears on control tasks with rare reward events.

arXiv:2610.29430v1 Announce Type: new Abstract: Multimodal multi-label emotion recognition is important for affect-aware systems, but models trained with complete text, visual, and acoustic streams often degrade sharply when some streams are unavailable at deployment. We address this train-test gap by learning a compact representation that preserves emotion-specific evidence from any available subset of modalities. We propose the Multi-modal Emotion Recognition Perceiver (MERP), which maps the joint sequence into a fixed-size latent bank, thereby keeping repeated reasoning independent of raw input length; it extracts label-specific evidence before and after latent reasoning and aligns the two levels through reconstruction. MERP further uses positive and negative class prototypes, a noise-gated feature enhancer with frequency-shaped perturbations during training, and confidence-aware class-wise fusion to strengthen separability, reduce sensitivity to missing inputs, and combine complementary predictors. Under 20 percent test-time random modality drop, MERP achieves micro-F1 scores of 0.563 on CMU-MOSEI and 0.780 on M3ED, exceeding the strongest evaluated baselines by 0.004 and 0.005, respectively; in the severe CMU-MOSEI audio-only setting, it retains 0.473 micro-F1 versus 0.022 for LDDU. These results show that fixed latent reasoning provides a practical, testable route to robust multimodal emotion recognition without requiring missing-modality examples during training. Code is available at https://github.com/Creative-zcx/MERP.

arXiv:2610.29414v1 Announce Type: new Abstract: Positional superko (PSK) forbids recreating any earlier whole-board position. It is part of the Chinese rules as written and of the rules used in computer Go. Yet almost all solved Go boards were computed under a simpler basic-ko rule set. PSK makes solving hard: legality depends on the whole game history, and this breaks the three tools that make solving tractable, namely transposition tables (the graph-history interaction problem), static life-and-death recognition, and safe move pruning. We present Sanko, a parallel depth-first proof-number solver that restores all three soundly. Its core is a cheap proof-transfer theorem: a stored proof stays valid on a new path unless a board that the proof's winner must create has already occurred. Sanko proves 5x5 Go in 16 minutes with six threads, where the previous state of the art needed 5.25 hours with six processes, and in about a minute and a half on one 192-thread machine. It also establishes the first PSK values of twenty-three further boards, eleven of which were unsolved under any rules. Superko changes game-theoretic values: 4x7, a full-board win under basic ko, is worth only 12 points under PSK, and we pinpoint the move at which the previously published optimal line fails. Unlike earlier solutions of empty Go boards, we release the solver source together with the command line and outcome of every run behind our values and measurements, so that all reported values can be reproduced.

arXiv:2610.29331v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly applied across the scientific discovery pipeline, from synthesizing literature to generating hypotheses to informing experimental design. A unifying principle behind these tasks is the need to continuously update beliefs as evidence accumulates, for which Bayesian inference provides a natural framework. However, existing work assumes that LLMs cannot carry out principled Bayesian belief updating in context, instead relying on simplifications and heuristics. To challenge this assumption, we present a principled evaluation of Bayesian reasoning capabilities in LLMs, analyzing both the model's internal consistency as well as its ability to recover an externally determined posterior. Specifically, we build an evaluation framework spanning qualitative epistemological examples to probe probabilistic coherence, quantitative benchmarks of Bayesian inference capabilities, and analyses of the impact of non-Bayesian updates on downstream task performance. Across different scenarios, we find that LLMs generally violate Bayes' rule when updating their beliefs in context. However, when equipped with modern reasoning abilities and code execution capabilities, their performance drastically improves, enabling even the close approximation of analytically intractable posteriors. Our evaluation shows that current LLMs with access to appropriate tools could already be used to make complex belief updates in scientific discovery, enabling rapid acceleration of the efficiency and scale of AI-powered scientific discovery.

arXiv:2610.29330v2 Announce Type: new Abstract: How do LLM agents' strategies differ, and how much can instructions to adopt a persona change their behavior over a game? We study these questions with AgentMonopoly, an environment where agents communicate, buy and trade properties, and make decisions that affect later turns. We compare models from four families, each at two sizes, in games with identical models, different sizes of the same family, or different families. We also assign one agent a persona, a description of how it should behave, intended to change a tendency observed in games without that persona. The tested models differ in how often they trade, waive rent, and manage cash and properties. Larger models do not consistently win more games. Persona instructions change the words agents use and some of their actions, but the changes do not always match the role across all decisions. These results show why persona following should be assessed through actions and outcomes as well as language. As our findings are conditional on the tested checkpoints, prompts, and five seeds per configuration, the open-source nature of AgentMonopoly provides the experimental infrastructure for further extending these comparisons. Our code is available at https://github.com/SuELS-Group/Agent_Monopoly.

arXiv:2610.29317v1 Announce Type: new Abstract: Spatial reasoning requires more than access to geometric tools: a reasoner must decide when direct observation is sufficient and when to reconstruct, measure, or change reference frames. We present SpatialBrain, a framework that retains experience about these decisions while keeping its vision-language backbone fixed. The key distinction is between evidence about a scene and experience about how to reason with that evidence. 3DFusionMemory fuses current-scene observations into queryable spatial relations. A Spatial Skill Library provides reusable tool procedures, and Spatial Cognition associates task conditions with experience about when and how to use memory and skills. Completed trajectories are abstracted into process records and consolidated between task blocks, allowing later decisions to draw on earlier execution without transferring scene coordinates or reference answers. Across five spatial-reasoning benchmarks, SpatialBrain improves on direct answers from the same backbone on four and ties on the fifth. Component comparisons, experience-coverage curves, and traced resource decisions characterize how the framework accumulates and uses these resources.

arXiv:2610.29308v1 Announce Type: new Abstract: Dynamic scheduling requires dispatching online arrivals while anticipating future jobs, as poor decisions can compound downstream costs under nonlinear tardiness objectives. We show that a pretrained large language model (LLM) can serve as the exogenous component of a structured world model, forecasting uncertain future job arrivals. Specifically, we exploit an afterstate decomposition, which separates known deterministic shop transitions from uncertain exogenous job arrivals, enabling the LLM to forecast only the latter through a compact belief representation. Building on this, we develop Afterstate World-Model Planning (AWMP), which combines an evolved, belief-conditioned dispatching strategy with short scenario-based rollouts. A surprise-triggered refresh mechanism queries the LLM only when newly observed arrivals deviate sufficiently from previous forecast, and online residual self-calibration further conditions the LLM on its recent forecast errors to handle nonstationary arrivals. Across diverse arrival distributions, AWMP reduces weighted tardiness by 6-21% on dynamic job-shop and 22-46% on flexible job-shop benchmarks, relative to the best of eight classical priority-dispatching rules.