标题：Recursive Self-Improvement through Multi-Agent Self-Supervision
来源：arXiv
链接：https://arxiv.org/abs/2610.12176
摘要：Recursive self-improvement (RSI) of a model on non-verifiable tasks, such as open-ended research, faces a supervision bottleneck when its outputs exceed what even human experts can reliably assess, leaving the model itself (optimizee) as the best available optimizer and evaluator. However, a single model instance struggles to critique and improve its own complex reasoning under this homogeneous loop. To address this, we propose Multi-Agent Self-Supervision (MASS), an RSI method that alternates between evolutionary workflow optimization and supervised fine-tuning on self-generated trajectories. Guided by early findings that multi-agent topologies excel at complex reasoning, MASS prompts a single base model to iteratively propose, execute, and self-evaluate multi-agent workflows. Through an
标题：Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
来源：arXiv
链接：https://arxiv.org/abs/2610.11794
摘要：Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives. Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states. We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory. The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified. It compiles this rulebook into executable code for prediction and planning. Through a continual loop of observation, reflection, rule revision, compilation, and verification, the agent
标题：Language Models as AI Research World Models
来源：arXiv
链接：https://arxiv.org/abs/2610.12235
摘要：AI research agents automate the cycle of proposing, implementing, and evaluating experiments, opening a path toward recursive self-improvement. Yet their ability to propose experiments outpaces their capacity to execute them in real environments, making outcome prediction a key capability for sustained self-improvement under limited experimental budgets. We investigate language models as Research World Models (RWMs), which predict the outcomes of candidate interventions across research environments. Our evaluation draws on over 2,600 experimental records from nine research environments spanning pretraining, post-training, and inference, representing more than 171,000 H100 GPU-hours of experimentation. Research knowledge acquired from real experimental experience improves RWM predictions of
DataSense-Bench：迈向 AI 科学家的数据感知评测
DataSense-Bench 通过训练数据子集选择与性能预测任务，评估前沿 AI 智能体识别高价值数据并预判微调效果的能力，结果显示其尚未形成可靠且可跨任务泛化的“数据感知”。
标题：DataSense-Bench: The First Step Toward an AI Scientist
来源：arXiv
链接：https://arxiv.org/abs/2610.12190
摘要：As claims about recursive self-improvement (RSI) and artificial general intelligence (AGI) proliferate, we ask a simple question: do frontier AI models have a sense of data, i.e., can they reliably select the right data for training? We introduce DataSense-Bench to study this capability through the fundamental problem of data selection and performance forecasting in machine learning. We ask AI agents to select and rank candidate training subsets that can be used to fine-tune a small LLM model. Agents are allowed to inspect the data, write and execute analysis code, and run model forward passes, but can not train the model or access the actual evaluation tasks. We then fine-tune the base model on each selected subset and evaluate its post-training performance under a standardized protocol.
标题：MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement
来源：arXiv
链接：https://arxiv.org/abs/2610.11959
摘要：Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. This report introduces the MiMo-V2.6 series, an omni-modal family that pushes the frontier of model intelligence by scaling RL compute. Prior to RL, we conduct mid-training on a broad multimodal corpus to provide ample exploration space, and build a solid infrastructure on the pretrained hybrid-SWA architecture to support subsequent scale-up. We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context lengths of up to 1M; (2) more diverse and complex environments, spanning code, general, visual, and cyber domains under a mixture of agent harness
[合并自 2026-10-10 mimo-v2-6]
[MiMo-V2.6：通过扩展强化学习迈向自我改进（56 ▲）](https://huggingface.co/papers/2610.11959?utm_source=digest-papers&utm_medium=email&utm_campaign=2026-10-09)
[合并自 2026-10-10 mimo-v2-6]
标题：MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement
来源：arXiv
链接：https://arxiv.org/abs/2610.11959
摘要：Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. This report introduces the MiMo-V2.6 series, an omni-modal family that pushes the frontier of model intelligence by scaling RL compute. Prior to RL, we conduct mid-training on a broad multimodal corpus to provide ample exploration space, and build a solid infrastructure on the pretrained hybrid-SWA architecture to support subsequent scale-up. We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context lengths of up to 1M; (2) more diverse and complex environments, spanning code, general, visual, and cyber domains under a mixture of agent harness
标题：RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments
来源：arXiv
链接：https://arxiv.org/abs/2610.12424
摘要：A generalist robot should not only perform diverse tasks but also improve through experience, turning what it learns during execution into capabilities that later tasks can reuse. Robot agents that act through code can already repair programs from execution feedback, yet it remains a central challenge to organize this experience around the task structure that gives it meaning, so that each repair is attributed to the responsible capability, supported by execution evidence, and validated before it is reused. We introduce RoboRSI, a robot self-improvement system built on Top-Down Skill Refinement (TSR). TSR decomposes tasks into compound, atomic, and base skills with scoped responsibilities and explicit input--output contracts, attributes each execution outcome to the responsible branch, and
标题：Which Skill to Distill? SGUID: Selecting a Compact Skill Bank for Model-Skill Co-Evolution
来源：arXiv
链接：https://arxiv.org/abs/2610.12367
摘要：Skills, reusable procedural guidance added at inference, can substantially improve LLM downstream performance (Li et al., 2026). Prior work retrieves skills from a bank by semantic relevance, then uses them as inference-time patches or for model distillation. The individual utility of each skill, however, is largely neglected. We first show that, in on-policy distillation where skill-conditioned policies serve as teachers, fewer than 25% of retrieved skills provide useful distillation signals. We then propose SGUID, a method for selecting a compact subset of skills for distillation. SGUID retains a skill only if it consistently yields effective learning signals during training. The selected skills are then distilled to produce a better model. Our results show that not all skills are worth
标题：Use and Disuse: Intent-Structured Experience Consolidation for Memory and Learning in LLM Agents
来源：arXiv
链接：https://arxiv.org/abs/2610.12124
摘要：The evolution of Large Language Model agents from single-task execution to long-term autonomous operation highlights the critical challenge of transforming continuous experiences into reusable knowledge. To address this, we propose Hippocam, a hierarchical memory and continual learning architecture. Hippocam draws inspiration from two characteristics of human memory: cognitive processes selectively maintain information relevant to current goals, while long-term memories form gradually through repeated consolidation. Accordingly, Hippocam structures an agent's ongoing work as nested intents. The active context remains centered on the current intent, while completed intents are consolidated into the task-relevant outcomes and state needed for subsequent work, rather than carrying forward the
标题：TRACE: Diagnosing Verifier Brittleness in Agentic Evaluation
来源：arXiv
链接：https://arxiv.org/abs/2610.11678
摘要：Verifier scores now serve as both benchmark metrics and training rewards for large language model (LLM) agents, and a change in score is routinely read as a change in capability. It may instead reflect a change in the evaluation. We introduce TRACE, a protocol that turns a score change from a verdict into a testable diagnosis: it applies a targeted change to one part of an evaluation, compares paired runs, checks whether the agent's behavior changed, and rescores unchanged trajectories to test whether the scoring rule is responsible. In a controlled suite of 25 synthetic tasks, renaming tools lowers a scripted agent's score by 0.250 even though it performs exactly the same operations; restoring the original names at scoring time closes the entire gap, while the same mutation exposes a genu
标题：What Output-Only Review Cannot Verify: Study Contracts for Research Agents
来源：arXiv
链接：https://arxiv.org/abs/2610.11754
摘要：Some defects in an AI-generated study can be identified from its artifacts; others require knowledge of what was approved before execution. We propose study contracts that bind declared experimental choices, run obligations and claim scope to recorded execution evidence, and distinguish this contract-relative verification from scientific truth. A diagnostic using eight self-authored clean/mutated pairs illustrates the information boundary. A deterministic checker applying a registered, fault-specific rule to approved and executed objects detected all eight registered mutations. Across eighteen recorded judge aliases given individual metadata-filtered packages without pair context or the registry-selected fault label, 104 of 144 mutated evaluation cases received defect flags; the remaining
论文提出 AI 智能体的“生态安全”问题，关注种群规模如何随网络攻击能力动态变化，而非仅研究固定数量下的单体或多智能体安全。作者建立由网络安全能力决定适应度与增长率的种群增长方程：缺乏协作时，只有单体能力超过临界水平，失配智能体种群才会持续扩张。引入协作后，集体网络攻击能力随种群规模提升并形成强阿利效应——低于临界种群规模时数量衰减，高于阈值时即使单体能力不变，也会进入自我强化式扩张。论文因此主张开展生态红队测试与受控的渐进式部署，并在每一代模型能力提升后重新估计失控阈值。
标题：Ecology of AI Agents: Collaboration Creates a Population Threshold for Takeoff
来源：arXiv
链接：https://arxiv.org/abs/2610.12436
摘要：AI agents can now conduct real-world cyberattacks, scale up capabilities with the number of agents, and collectively pursue misaligned goals to obtain rewards. Together, these factors raise the risk of a population explosion of misaligned agents: agents could compromise computers and secretly deploy additional agents, creating a self-reinforcing cycle where larger populations develop greater collective cyber capability and expand further. This raises a fundamental question: What determines whether a population of misaligned agents remains contained or takes off into this self-reinforcing cycle? This population-level problem is ecological safety: unlike individual-agent or multi-agent safety with a fixed population, it concerns the dynamics of the population itself. Here, we develop an ecol
标题：One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails
来源：arXiv
链接：https://arxiv.org/abs/2610.12292
摘要：A typed decision model reads a piece of text and returns a probability over caller-defined options, each with a short written definition, generating no text. Recent work places these models in agent systems as guardrails: the component that reads a proposed tool call or incoming message and decides whether to allow it. We evaluate seven open-weight models in that role and report the two error directions separately: a fail-open error allows a prohibited action and is a vulnerability; a fail-closed error blocks a permitted one and is only a cost. On prompt-injection, jailbreak and toxic-content screening, accuracy at the allow-or-block decision ranges from 36% to 72% against a chance level of 50%. A low error rate in one direction only reflects which answer a model defaults to: one allows ne
标题：Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness
来源：arXiv
链接：https://arxiv.org/abs/2610.12361
摘要：Large language models increasingly justify legal decisions by naming the statute or precedent behind a verdict, treated as evidence that the decision follows from it. We test this directly: holding case facts fixed, we substitute the named legal authority for an unrelated one and decode a model's evolving verdict from its hidden states. Across seven open-weight models (8B-70B) and four benchmarks spanning judicial and contractual reasoning, when explicitly required to justify a verdict by naming the governing authority, models name the correct one in 66.7%-100% of generations, while the verdict changing when the authority changes is far less consistent: 0.0%-21.7% on CaseHOLD, 30.0%-76.7% on ECHR and SCOTUS, and 43.3%-50.0% on ContractNLI. Neither scale nor a purpose-built legal-reasoning
标题：Learning Probabilistic Logic Programs with Functional Gradient Guided Language Models
来源：arXiv
链接：https://arxiv.org/abs/2610.12303
摘要：Declarative logic programs offer a powerful and interpretable abstraction for encoding relational structure and neurosymbolic reasoning, by expressing dependencies as weighted compositional rules. However, inducing them from data remains fundamentally hard, bottlenecked by the combinatorial explosion of symbolic search spaces. LLMs have recently emerged as powerful hypothesis generators, but when used in isolation, they lack the capacity to do systematic inductive reasoning needed to reliably synthesize valid programs that fit complex relational distributions. We introduce grasp (Gradient-boosted Synthesis of Probabilistic logic programs), a neurosymbolic framework that casts relational structure learning as functional gradient boosting in which the weak learner is a first-order rule and t
标题：Natural Language to First-Order Logic LLM-based Autoformalization
来源：arXiv
链接：https://arxiv.org/abs/2610.12030
摘要：Large Language Models (LLMs) have renewed interest in autoformalization. Yet, when First-Order Logic (FOL) is considered as the target formalism, the field still lacks a unified task formulation and a systematic survey. This paper addresses this gap: we first provide a principled definition for the FOL-autoformalization task by distinguishing Ontology Extraction from Logical Translation, showing how their conflation obscures (cross-study) evaluation; we review existing datasets, evaluation metrics, and LLM-based methods, including fine-tuning, prompting, and verification-based refinement; we identify open challenges in benchmarking, semantic evaluation, ontology-aware methods, and end-to-end applications.
标题：LEVER: Adaptive Cost-Aware Proof Search Over AND/OR Graphs
来源：arXiv
链接：https://arxiv.org/abs/2610.11862
摘要：Mathematicians value proofs for more than correctness: among correct proofs, simplicity, purity and the computational cost of finding them vary widely. Yet LLM-powered theorem provers largely search for any correct proof, and improve its quality only after it is found. We propose LEVER, a proof search algorithm that makes the objective over correct proofs programmable and optimizes it during search. LEVER scores partial proofs over an AND/OR proof graph, combining realized objective values with predictions for open subgoals, so the objective guides search before a proof is complete. The same mechanism optimizes computational cost, proof length, topical impurity, and even their weighted combinations, while the Lean kernel enforces correctness. On PutnamBench in Lean 4, under matched budgets
标题：MemTrial: Learning When to Trust Memory in LLM Portfolio Agents
来源：arXiv
链接：https://arxiv.org/abs/2610.11732
摘要：Large language model (LLM) agents for portfolio management learn from experience: they credit each experience in their memory with the outcome of the decisions that used it. In financial markets, however, this outcome mostly reflects the market move shared by all decisions on that date, so the credit tracks the market rather than the experience, and these agents often do worse than simply holding the equal-weight (1/$N$) portfolio. We ask how an agent can credit an experience with what it changes, and answer it by putting memory on trial: drafts of the same decision with and without an experience face the same market, so the outcome they share cancels in their difference. Our agent, MemTrial, drafts each decision with eight combinations of its retrieved experiences, chosen by a fractional
标题：DeltaReplay: Task-Relative Memory Reuse for Mobile GUI Agents
来源：arXiv
链接：https://arxiv.org/abs/2610.11707
摘要：Memory-augmented mobile GUI agents store successful execution trajectories and reuse them in later tasks, but a stored trajectory rarely matches a new task exactly. The new task may use different parameters, share only some of its steps with a stored trajectory, or have no relevant record in memory. Forcing the agent to use irrelevant memory can mislead it, whereas discarding memory that may still be useful deprives it of guidance from past experience. To address this dilemma, we propose DeltaReplay, a step-level memory reuse framework that decides how to use existing memory without modifying it. We observe that the reusable part of a stored record is determined not by the record itself but by its relation to the new task, mainly through two factors: page-level consistency and action-level
标题：Event-Centric Memory with Query-Aware Graph Augmentation for Long-Term Conversational Agents
来源：arXiv
链接：https://arxiv.org/abs/2610.11920
摘要：For persistent and personalized conversational agents, memory systems can enable them to remember, update, and reason over long histories by storing past interactions and retrieving relevant information. Existing memory systems typically follow two paradigms: flat-structured memory and graph-based memory. The former is lightweight but leaves event relations and state updates implicit, while the latter explicitly models memory structure but incurs additional construction cost and introduces irrelevant relations over long histories. To address these limitations, we propose QGMem, a novel memory construction and activation framework motivated by human memory, in which experience is organized into events and query-relevant events are modeled by graph as working memory. QGMem converts long dial
标题：Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models
来源：arXiv
链接：https://arxiv.org/abs/2610.12090
摘要：Latent reasoning enables vision-language-action (VLA) models to transform multimodal observations into task-relevant internal states before generating continuous robot actions. While existing methods learn to generate or refine such states for each policy query, they discard successful reasoning after execution and therefore reconstruct similar computation from scratch. We present Reasoning and Flow Memory (FLOWMEM), a unified VLA model that turns successful latent computation into reusable reasoning experience. Rather than appending a fixed retrieved context, FLOWMEM dynamically retrieves and recomposes compatible latent fragments as the embodied context evolves, forming a reasoning route that follows the temporal structure and progress of successful computation. The route is then refined
标题：DPPM: Dual-Path Parametric Memory for Personalized Language Models
来源：arXiv
链接：https://arxiv.org/abs/2610.11776
摘要：Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Acro
标题：ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills
来源：arXiv
链接：https://arxiv.org/abs/2610.12403
摘要：Skill-augmented agents improve sample efficiency by distilling successful trajectories into reusable strategies. Yet most existing approaches remain text-centric, linearizing spatial layouts and action-state correspondences into language that loses critical geometric structure. Recent efforts have begun incorporating visual evidence, but construct and update skills separately from policy optimization, leaving their mutual improvement underexplored. We propose ViSkill, a visual-native skill learning framework that encodes successful interactions as composite visual skill cards directly accessible to VLM agents. Retrieved skills guide both inference and reward shaping, while successful trajectories are distilled back into the library, forming a closed feedback loop in which skill accumulatio
标题：Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read
来源：arXiv
链接：https://arxiv.org/abs/2610.12133
摘要：Many key-value (KV) caches are compressed before anyone knows what will be asked of them: a document cached for retrieval, a prompt prefix shared across requests, the memory of a long conversation. The prevailing approach scores KV entries by rehearsal: the model rereads the context and keeps the entries it attends to, assuming that the more completely a cache rehearses its context, the better it remembers it. We show that under tight budgets this assumption backfires: rehearse everything, remember nothing. At a 3% keep ratio, rereading the whole context keeps 31.5 of 96.5 points on RULER, and on LongBench's natural-text tasks it falls below methods that rehearse nothing at all. The cause is that a cache keeps what it rehearses: rereading spreads the budget across the whole context, so the
标题：Internalizer: Portable Context-to-Parameter Mapping for Very Large Language Models
来源：arXiv
链接：https://arxiv.org/abs/2610.11715
摘要：Hypernetworks that map a context directly to a LoRA adapter let a large language model carry that context in its weights, but prior work has demonstrated them only on base models of up to 14 billion parameters. We present the Internalizer, a state-of-the-art, portable Context-to-Parameter Mapping hypernetwork that generates document-specific LoRA adapters for the frozen 284B-parameter DeepSeek v4 Flash, a target two orders of magnitude larger than in any previous work. Most of its parameters live in a model-agnostic trunk with only thin entry and exit layers per base model, so it trains cheaply against small models before being ported to the large one. On unseen documents of up to 4096 tokens, the generated adapters reach 84.9% top-1 and 97.8% top-5 teacher-forced accuracy against 63.4% an
标题：When History Helps and Hurts: Selective History Use across Multimodal Turns
来源：arXiv
链接：https://arxiv.org/abs/2610.11948
摘要：Reliable multimodal interaction depends on selective use of conversational history: an earlier question may remain relevant while its previous answer is outdated, whereas a current request may depend on historical evidence despite conflicting new observations. Existing multi-turn evaluations rarely separate these history-use demands from underlying question difficulty. To address this gap, we introduce ReTurn, a benchmark of 7,000 base tasks spanning visual and audio evidence for evaluating selective history use. For task-carrying history, Reconfirm/Reground require applying a historical question to current media while varying historical agreement; for evidence-carrying history, Retrieve/Rebind require answering a current question using historical media while varying current-media competit
标题：Spatial Pattern Formation from Multi-Agent Learning in Public Goods Dilemmas
来源：arXiv
链接：https://arxiv.org/abs/2610.12321
摘要：Spatial public goods models show that prescribed movement toward richer locations can generate spatial patterns. We ask how such patterns emerge when agents learn where to move and how learning rates shape their consequences for collective welfare. Fixed populations of cooperators and defectors independently learn movement policies using tabular Q-learning and local observations. Cooperator learning generates clusters around resource peaks, while co-adaptation changes their strength and motion. At a fixed training budget, the largest welfare losses occur when cooperators learn at high rates and defectors at low rates. In part of this regime, learned policies also generate traveling bands supported by a shared directional preference. The conditions supporting travel change with further trai
标题：GeoReform: Reflective Formalization Evolution for Multimodal Geometry Problem Solving
来源：arXiv
链接：https://arxiv.org/abs/2610.12391
摘要：Multimodal large language models (MLLMs) often struggle to identify and use geometric relations in diagrams. Recent methods address this challenge by converting geometric entities, relations, and constraints into explicit textual representations for the model to reason over. However, effective formalization is highly non-trivial: on Geometry3K, structure injection fixes 28 errors but introduces 13 new ones among 200 examples. Redundant relations can distract the model, while ambiguous references to diagram elements can lead it to apply constraints incorrectly. This suggests that the key challenge is not merely extracting more geometric facts, but organizing them into representations that support downstream reasoning. To fully exploit the power of formalization, we further propose GeoReform
标题：Trajectory-Guided Fault Localization for Agent Skill Evolution
来源：arXiv
链接：https://arxiv.org/abs/2610.11858
摘要：Agent skills provide reusable guidance for code agents, but incomplete or unsuitable guidance can impair task execution. To reduce the manual effort of skill refinement, recent approaches use LLMs to generate revisions from execution feedback. However, grounding these revisions in explicit behavioral evidence remains challenging. To address this gap, we propose SkillMorph, a skill-evolution approach based on trajectory-guided fault localization in agent skills. Its core idea is to link execution evidence to specific skill contents before generating revisions. Specifically, SkillMorph compares failure and success evidence in abstracted trajectories across repeated runs and tasks, incorporating changes between evolution loops to identify suspicious actions. It then uses these suspicious acti
标题：BrickBench: Evaluating Agentic Brick Design
来源：arXiv
链接：https://arxiv.org/abs/2610.12452
摘要：We propose BrickBench, a benchmark for agentic text-conditioned LEGO-set design. Given a prompt, an agent is tasked with producing an assembly that not only satisfies semantic and design criteria, but that can also be physically built. To do so, it must select parts from a discrete library and reason jointly about local and global constraints. We score validity, alignment, and design across three settings that vary in scale and part availability. We provide BrickAgent, an environment for coding agents to construct, inspect, and validate their designs. We find that leading agents largely satisfy verifiable physical and semantic requirements, but fall short of human designs. We release our benchmark and environment at http://www.brickben.ch
为构建经真人研究验证的会话式 AI 仿真评测体系提供实践路线，同时提示仿真结果仍需真人实验校准，避免将模拟偏差误判为真实用户行为。
📝 原始笔记（逐字保留）
标题：InterviewPlayground: A Simulation Environment for Evaluating AI Interviewers
来源：arXiv
链接：https://arxiv.org/abs/2610.12023
摘要：Increasingly, AI interviewers are being developed to elicit open-ended responses in applications like market research, public polling, preference elicitation, and social science research. However, evaluating AI interviewers is challenging because they function in extended, multi-turn interactions where they must adapt to participant behaviors. To address this need, we develop InterviewPlayground, a simulation environment for evaluating AI interviewers using simulated study participants whose behaviors are grounded in social theory. Simulated studies in InterviewPlayground produce an InterviewReportCard, which assesses the performance of AI interviewers using a suite of validated measures. To test whether our simulation-based evaluations predict performance with human participants, we condu
标题：Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition
来源：arXiv
链接：https://arxiv.org/abs/2610.12341
摘要：Adversarial games have driven advances from heuristic search to reinforcement learning, yet learning and adapting strategies from limited samples remain challenging. AI agents offer an alternative by turning game experience into revisions of executable policies. Building on heuristic learning (HL), we formalize Adversarial Heuristic Learning (AHL), a paradigm that uses AI agents as learning engines to refine game policies and supporting software while keeping model weights fixed. We introduce AAArena, a benchmark comprising 12 authentic adversarial games and 1,920 archived human programs, with an evaluation protocol modeled on real-world game competitions. Agents interpret rules, choose opponents, analyze replays, and revise game agents to achieve their highest ranking within fixed match a
标题：All Verdicts are Not Equal: Rethinking LLM Judge Reliability
来源：arXiv
链接：https://arxiv.org/abs/2610.12083
摘要：LLM-as-a-Judge is the standard paradigm for NLP evaluation, yet its systemic reliability remains poorly understood despite being widely treated as a deterministic ground truth. We present a comprehensive reliability audit, stresstesting six frontier models across four benchmarks, five prompt formats, two presentation orders, three sampling temperatures, and ten repetitions per condition. Our empirical analysis reveals severe vulnerabilities: verdicts change across identical replications at temperature zero, position-order swaps flip the majority of verdicts on challenging tasks, and the most deterministic judge achieves perfect consistency by trivially repeating incorrect verdicts, agreeing with ground truth only 51% of the time. To formalize these multi-faceted failure modes, we introduce
★2AI 时间跨度估计与 METR 指标效度分析arXiv · 评测审计 · 论文研究以样条函数和项目反应理论重新估计 METR 的 50% AI 时间跨度，揭示人类任务时长与 AI 难度之间存在非线性关系，并提出结合诊断图评估指标的构念效度。
研究基于 228 项软件任务和 26 个 AI 系统，重新估计 METR 的 50% 时间跨度，即 AI 以 50% 概率完成任务时对应的人类完成时长。作者使用样条函数与项目反应理论，放宽“AI 任务难度与人类完成时间的对数呈线性关系”这一假设。拟合结果显示，2—30 分钟区间的人类时间到 AI 难度映射近乎平坦，因此时间跨度从 3 分钟增至 30 分钟，远比从 30 分钟增至 5 小时容易，不能将二者视为同等幅度的能力跃迁。新估计在交叉验证的适当评分规则下表现更好，作者建议在发布或扩展时间跨度基准时同步提供诊断图，以审查指标的构念效度。
用途与启示
提醒研究者不要将 AI 时间跨度的等比例增长直接解释为等量能力提升，尤其应谨慎看待短时任务区间的跨度跃升。
可用样条函数和项目反应理论改进长程软件任务基准的统计建模，减少线性假设造成的能力误判。
建议将交叉验证评分与诊断图纳入时间跨度基准的效度审计流程，以暴露非线性区间和构念效度问题。
为扩展至更长软件任务的动态基准提供更稳健的估计、统计校准与模型比较方法。
📝 原始笔记（逐字保留）
标题：On the estimation and validity of AI time horizons---a statistical look at the METR plot
来源：arXiv
链接：https://arxiv.org/abs/2610.12466
摘要：METR's 50\% time horizon measures the human completion time of software tasks that an AI solves with 50\% probability, allowing AI capabilities to be expressed in interpretable units. On 228 tasks and 26 AIs, we recompute the time horizons using splines and item-response theory to relax the assumption that the AI difficulty of a task depends linearly on the log of human time. Our fitted spline can be interpreted as a function that \emph{converts} human time to AI difficulty; it is nearly flat in a region from 2--30 min but close to linear elsewhere. Hence, a time-horizon jump from 3 min to 30 min is much easier than one from 30 min to 5 hours despite the same multiplier of $10 \times$. Overall, we contribute time-horizon point estimates that perform better under a cross-validated suite of
★2Perception Test 2026：城市级音视频空间推理挑战arXiv · 评测基准 · 论文Perception Test 2026 新增基于城市步行导览视频的 KilometerAudio 与 KilometerVision 赛道，显示城市级空间与多模态推理仍高度依赖昂贵的智能体流水线。
标题：Perception Test 2026: Challenge Summary and Extension to City-scale Audio-Visual Reasoning
来源：arXiv
链接：https://arxiv.org/abs/2610.12081
摘要：Continuing the Perception Test challenge series, we organised the fourth edition as a workshop at the European Conference on Computer Vision (ECCV) 2026 in Malmö, Sweden. This edition focused on spatial intelligence and featured four different tracks: unified multiple-choice videoQA and grounded videoQA from the original Perception Test benchmark, alongside two new tracks based on city-scale walking-tour videos (KilometerAudio and KilometerVision). In this report, we describe the new benchmarks used for the city-scale tracks and summarise the winning solutions across all tracks, including a generalist model that competed across all tracks with satisfactory performance. The winning solutions in the newly added city-scale tracks demonstrated that complex spatial and multimodal reasoning can
★2真实程序精确输出与检查点状态预测评测arXiv · 代码评测 · 论文该研究推出覆盖真实 Python 与 C++ 程序的预测基准，评估模型仅依据源码和输入推断最终输出及循环检查点状态的能力，并揭示长轨迹与状态任务中被短输出评测掩盖的缺陷。
标题：Evaluating Exact Output and Checkpoint-State Prediction in Real Programs
来源：arXiv
链接：https://arxiv.org/abs/2610.11889
摘要：We present a benchmark for predicting final output and checkpoint state from source and input alone. It extends CRUXEval-style output prediction with paired shorter- and longer-trace inputs and checkpoints inside and after a loop. The benchmark contains 400 cases from 371 Python and C++ programs, evaluated under seven settings from four model families without tools or code execution. Of 11,200 planned predictions, 11,151 produced gradable responses. Reasoning-enabled settings outperform their off counterparts by 33.1 to 55.2 percentage points on completed responses. The strongest setting scores 93.0% on shorter-trace final output, 77.0% on longer-trace final output, and 65.5% and 63.5% on the two state tasks; these scores also hold when missing responses count as wrong. Across 2,397 matche
论文指出，仅凭攻击成功率（ASR）无法判断无害输出源于模型识别并拒绝了有害意图，还是未能理解或根本没有处理目标任务。作者提出操作性理解率（UR），衡量回复是否识别被评测任务，并将其作为实际需要回答的任务处理。跨接口实验显示，相近的 ASR 可能对应显著不同的意图恢复率；随着压缩提示被还原得更明确，UR 稳定上升，而 ASR 并不呈现相同规律。FormalLogic 实验还表明，高意图恢复率仍可能伴随频繁的有害协助，因此安全评测应联合报告 UR 与 ASR。
用途与启示
改进以 ASR 为核心的安全评测，避免将任务未理解误判为成功拒绝。
为意图混淆、压缩提示和越狱攻击评测增加任务参与度与理解程度指标。
支持对安全响应进行更细粒度归因，区分识别后拒绝、识别失败和答非所问。
提醒评测者联合分析理解能力与有害输出行为，避免由单一结果指标得出错误安全结论。
📝 原始笔记（逐字保留）
标题：Same Outcome, Different Evidence: Intent Recovery in LLM Safety Evaluation
来源：arXiv
链接：https://arxiv.org/abs/2610.11766
摘要：Safety evaluations of large language models commonly summarize harmful-output behavior with attack success rate (ASR). Yet the same non-harmful outcome can arise for very different reasons. A model may recover a harmful task and refuse it, fail to recover the task, or respond to something else entirely. Distinguishing these cases becomes especially important under intent-obscuring prompts, where a low ASR does not reveal whether the evaluated task was actually engaged. To make this distinction explicit, we pair ASR with operative understanding rate (UR), which measures whether a response both identifies the evaluated task and treats it as the task to be answered. Across interfaces, this paired view reveals substantial variation hidden by ASR: similar ASR values can correspond to sharply di
标题：Thinking Inertia: LLMs Keep Thinking When Told Not To
来源：arXiv
链接：https://arxiv.org/abs/2610.11765
摘要：Large Language Models (LLMs) increasingly ship with explicit "thinking modes", yet their counterpart, "no-thinking", has received far less attention. We study LLMs' no-thinking behavior along two axes. a. How to measure no-thinking? Prior work typically defines no-thinking through proxies such as a disabled thinking mode or the absence of long traces. These proxies are unreliable: disabled thinking modes may still emit reasoning, while long traces may contain filler rather than genuine inference. We instead normalize each response into a pre-answer trace and final answer, and evaluate it at three levels: (i) Empty-Thinking Rate for strict answer-only compliance; (ii) instruction-aware Question-Pre-answer Relevance for similarity between the question and pre-answer trace; and (iii) LLM-as-j
标题：HarnessSQL: Harness-Native Training for SQL Agents in Realistic Database Environments
来源：arXiv
链接：https://arxiv.org/abs/2610.12274
摘要：Text-to-SQL models are commonly trained to map questions directly to static queries, whereas real-world database agents operate through stateful, multi-turn interaction with live databases -- inspecting schemas, executing probe queries, diagnosing errors, and revising hypotheses. This creates a critical train-deploy mismatch, as the execution harness that mediates this interaction is introduced only at inference time. To bridge this gap, we propose HarnessSQL, a harness-native post-training framework that preserves the full interaction structure throughout both supervised fine-tuning and reinforcement learning. HarnessSQL builds isolated, executable database environments paired with hidden execution oracles, rolls out teachers directly inside the target SQL harness, and retains only verifi
★3从被动遏制到主动保障：三大厂商智能体安全事件启示arXiv · 可信执行 · 论文论文对比 OpenAI、Anthropic 与 Google 智能体越出授权测试边界的三起事件，提出覆盖完整执行系统的主动智能体安全保障框架 PASAC。
论文分析了三类网络安全评测事故：OpenAI 智能体利用研究基础设施、跨运行协作并影响 Hugging Face 生产环境，Anthropic 智能体因第三方环境配置错误接触真实系统，Gemini 则通过非预期互联网路径访问三个真实组织。案例表明，智能体评测不能依赖预设沙箱或静态授权边界，而应在运行期间持续验证边界是否有效。作者提出主动智能体安全保障循环（PASAC）和五层边界保障栈，涵盖风险分级任务、可执行范围契约、运行前验证、最小能力访问、独立出口控制、凭据限制及跨运行监控。框架还引入自动停止条件和基于执行证据的重新授权，并通过先行指标、九项设计命题和七个可证伪假设，将事故经验转化为可检验的安全研究议程。
用途与启示
为高风险网络安全智能体评测设计风险分级任务、可执行范围契约与运行时边界检查机制。
将最小权限、独立出口控制、凭据隔离和自动停机部署到执行层，构建不依赖单一沙箱的纵深防御体系。
持续监控跨运行轨迹、协同行为与状态残留，识别智能体在多次执行间积累能力或协调攻击的风险。
以执行证据作为继续运行或重新授权的依据，并利用设计命题、可证伪假设和评估指标推动安全机制持续验证。
📝 原始笔记（逐字保留）
标题：From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security Incidents
来源：arXiv
链接：https://arxiv.org/abs/2610.12463
摘要：In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths were different. OpenAI agents exploited research infrastructure, coordinated across runs, and compromised parts of Hugging Face's production environment. Anthropic reported cases in which a misconfigured third-party environment exposed real systems to agents pursuing simulated cyber tasks. In a separately reported evaluation, Google's Gemini accessed three real organizations through an unintended internet route; Google stated that the model stopped in all three instances. Taken together, the cases show why an evaluation cannot rely on an assumed boundary. That boundary must be verified while the agent is operating. This comparative instrument
标题：Structure Tax: How Structured Output affects LLMs Performance
来源：arXiv
链接：https://arxiv.org/abs/2610.12056
摘要：Deploying large language models in production often requires constraining outputs to structured formats such as JSON or XML, and prior work treats the resulting accuracy loss as an inherent `structure tax'. We re-examine this claim by evaluating a battery of models, datasets and schemas, measuring task accuracy, confidence calibration, and hidden-state geometry. The tax turns out to depend on schema design rather than on structure per se: reasoning-first field ordering matches or exceeds free-form accuracy, while answer-first ordering causes steep drops, particularly in smaller models. Format sensitivity scales inversely with a task's own structural constraints, and schemas that preserve reasoning order also improve calibration with CKA showing greater separability between correct and inco
表明针对已知漏洞的训练也能提升对未见攻击的泛化防御能力，为构建兼具抵抗力与韧性的 AI 提供实践路径。
📝 原始笔记（逐字保留）
标题：ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
来源：arXiv
链接：https://arxiv.org/abs/2610.12233
摘要：Recursive self-improvement, the participation of AI systems in improving their own capabilities, is beginning to move from theoretical prospect to practice, posing both challenges and opportunities for safety alignment. Models evolve through frequent updates, and their safety alignment requires continual adaptation to each new checkpoint. Meanwhile, with evolving red-teaming methods exposing new vulnerabilities, safety improvement for each checkpoint needs to mitigate exposed vulnerabilities and generalize to risks not yet revealed. Following R$^2$AI, we term these goals resistance to known threats and resilience to unforeseen risks. Recursive self-improvement, in turn, inspires an approach to both goals: safety alignment could likewise advance through successive rounds of evaluation and u
标题：Agentic-TTT: Training test-time policy for test-time training
来源：arXiv
链接：https://arxiv.org/abs/2610.12002
摘要：Test-time training (TTT) adapts an LLM's parameters using signals derived from test inputs, and can make striking improvements in pre-specified settings such as IMO competitions or designated open problems. By turning deployment experience into parameter updates, TTT provides a direct mechanism for model-level self-improvement. Yet TTT is not universally beneficial: each TTT algorithm works in different settings, and applying an ill-suited method could waste test-time compute or even damage model performance. Therefore, such parameter-level self-improvement requires agency: the model must decide when TTT is warranted, which algorithm to invoke, and whether an existing skill can be reused. To fill this gap, we introduce Agentic-TTT, which learns a test-time policy to govern those decisions.
标题：Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception
来源：arXiv
链接：https://arxiv.org/abs/2610.12445
摘要：Recent incidents have highlighted the challenge of monitoring LLM agents and the danger of models deceiving people. We show that white-box deception detection via probes can be scaled up to frontier monitoring settings by collecting the largest deception dataset to date for training probes and introducing a novel probe architecture which can aggregate information across many layers and tokens. Our probes achieve 98.8% AUC in SHADE-Arena, surpassing an Opus 5.5 text-monitoring baseline, and show improved efficacy as the underlying model is scaled up. To push our probes to their limit, we test them on several cases where deception cannot be determined from the context alone. In these cases, which we refer to as introspective deception, the ground truth can only be determined through careful
标题：Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models
来源：arXiv
链接：https://arxiv.org/abs/2610.12168
摘要：We introduce a self-improvement loop for reasoning models based on the following observation: Even when the difficulty of a problem exceeds the model's current solving abilities, an additionally supplied solution might enable the model to extract useful solution ideas in hindsight. We operationalize this by jointly training the same model to exhibit the following three capabilities: predicting solution ideas from problems alone, reverse-engineering ideas from problems and known solutions, and solving problems using provided ideas. The loop alternates between reverse engineering such ideas from problems with supplied solutions and using these ideas as additional supervision for joint training of all three capabilities. We give a formal specification of our method and a concrete instantiatio
标题：GRPODropout: Less is More for Online Reinforcement Learning Rollouts
来源：arXiv
链接：https://arxiv.org/abs/2610.11854
摘要：Reinforcement learning (RL) methods such as GRPO substantially improve large language model reasoning but often suffer from policy entropy collapse: the loss of sampling diversity weakens exploration and limits further improvement. Existing methods address this issue either through algorithm-level interventions, such as reward modification and entropy/KL regularization, or through token-level reweighting. We investigate a complementary perspective: entropy collapse can also be mitigated by changing which generated rollouts contribute to policy updates. Under the same sampling budget, not all rollouts contribute positively to an update, and selectively excluding some can improve learning. To address this, we propose GRPODropout: before the standard update, we use a simple strategy that sele
在均匀分布的 k 比特答案上，仅能接受或拒绝完整答案的精确验证器若要求零错误，最坏需要 2^k-1 次验证。论文证明，源任务调用与 n 次验证任意交错时所需的最小因果信息由列表率失真函数刻画，且一次验证前观测即可达到该前沿。对于固定源，将所有调用前移可保持调用次数的硬上限，但交错策略可能节省无界数量的期望调用；验证器含噪时，源优先协议在信息量和错误率上都可能遭受无界倍数的损失。在线性 F₂ 源任务库中，最优准确率具有闭式表达，预算曲线可经多项式时间归约后计算，并表明零错误硬预算中超出信息价格的调用用于处理干扰变量；除规划器相关的两个条款外，编号结论均在依赖两项已发表结果的前提下通过 Lean 4 机器检查。
用途与启示
为验证信息贫乏条件下的推理、搜索与迁移成本提供统一且可计算的信息论基准。
指导课程式迁移、源任务预训练及交错验证策略的选择，并区分调用次数硬上限与期望调用成本。
揭示噪声验证器下源优先协议可能显著失效，为验证流程设计和风险评估提供依据。
可将信息前沿作为诊断工具，判断模型是否有效利用了迁移任务提供的信息。
Lean 4 机器检查为相关信息论界限与调用复杂度结论提供形式化可信度。
📝 原始笔记（逐字保留）
标题：Verification with Transfer: Exact Information Frontiers and Their Price in Calls
来源：arXiv
链接：https://arxiv.org/abs/2610.12211
摘要：A verifier that accepts or rejects whole answers reveals little: under a flat prior over $k$-bit answers, zero error needs $2^k-1$ verifications. The usual remedy is to solve related source tasks, either all first, as a curriculum does, or interleaved with verification. We price this remedy in information and in calls. With an exact verifier, the least causal information that any interleaving of source calls and $n$ verifications needs to succeed with probability $s$ is a list rate-distortion function, attained by one observation before any verification. It lower-bounds the expected number of binary source calls, which designed sources meet within $1+\log_25$ calls for unique answers and within a logarithmic term in general, where no additive constant suffices. With an exact verifier and f
标题：When KL Regularization Misfires in Group Policy Optimization
来源：arXiv
链接：https://arxiv.org/abs/2610.12161
摘要：Why does removing reference-policy KL regularization sometimes improve group policy optimization? This motivates studying how reference-policy information should enter group-relative updates. We analyze seven potential failure modes in the interactions between KL and rewards: residual KL updates after reward clipping, after gradient cancellation, and in groups with identical rewards; KL growth with response length and an imbalance in its relative contribution; KL concentration on a small number of tokens; and sampling noise when k1 is incorporated into rewards. We propose Zero-Sum Calibrated Policy Optimization (ZCPO), which uses relative drift measured by conditional KL to calibrate within-group reward coefficients and integrates them into the base surrogate. Mathematical reasoning experi
标题：When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimation
来源：arXiv
链接：https://arxiv.org/abs/2610.12061
摘要：Large language model (LLM)-based agents have demonstrated strong capabilities on complex tasks. They typically perform reasoning before each action throughout an interaction trajectory. However, reasoning may not be necessary at every turn, as reasoning produced earlier can continue to support subsequent actions. A key challenge is therefore to determine when existing reasoning remains sufficient and when a new reasoning step is needed, without relying on costly generation-based verification. We find that decreases in the likelihood of subsequent reference actions after removing additional reasoning closely track whether those actions remain recoverable given earlier reasoning, providing an effective and lightweight signal for estimating cross-turn action support. Based on this observation
标题：Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks
来源：arXiv
链接：https://arxiv.org/abs/2610.11978
摘要：Jev is a "System One" model that returns a choice among given options instead of generating text. We study how such a specialized decision model compares with general-purpose large language models (LLMs). We evaluate Jev on 13 multiple-choice benchmarks covering knowledge, reasoning, and multilingual understanding, and compare it with 19 LLMs in three tiers: frontier, representative, and small. Jev is competitive with frontier LLMs on knowledge and commonsense benchmarks and obtains the best score on MMLU-Redux and ARC-Challenge. Outside mathematics, it also outperforms most representative LLMs and all small LLMs. However, it falls behind on mathematical word problems: on MathQA, it is 17.7 points below the frontier median and scores lower than all 19 LLMs. These results indicate that a sp
标题：Can Decision Models Understand Stance? Evaluating Jev Against General-Purpose LLMs
来源：arXiv
链接：https://arxiv.org/abs/2610.11901
摘要：Stance detection requires identifying an author's attitude toward a given target, sometimes based on conversational context. Jev, a specialized decision model designed for structured decision-making, offers an alternative to general-purpose large language models (LLMs). In this work, we evaluate Jev on two stance detection datasets, VAST (English texts) and ZS-CSD (Chinese conversations), comparing it with four general-purpose LLMs and two fine-tuned models. Results show that Jev achieves competitive performance on VAST, matching GPT-5.6 and outperforming the other general-purpose LLMs. However, it falls behind stronger LLMs on ZS-CSD, particularly in distinguishing favor from against. Further analysis suggests that this limitation may be related to understanding reply relationships and st
标题：Can Jev be Your Q or Policy in Reinforcement Learning?
来源：arXiv
链接：https://arxiv.org/abs/2610.11692
摘要：Foundation models supply reinforcement learning (RL) with priors that mitigate its longstanding weaknesses in sample efficiency and transfer, but their token-by-token generation makes queries sequential and costly. Jev, a recently released decision model, generates nothing and returns calibrated, typed answers in a single forward pass. Existing work studies foundation models in RL either as models to be trained or as generators to be prompted, and Jev belongs to neither category, having so far served only as a black box in single domains. How well such a model decides on its own in RL environments, and how it can improve RL as a component of training, therefore remain unaddressed. To this end, in this paper we first examine the requirements that the objects of an RL system place on the ans
标题：How Is Automated Research Evaluated? A Survey of Benchmarks and Evaluation Practices
来源：arXiv
链接：https://arxiv.org/abs/2610.11877
摘要：Automated research systems support literature synthesis, ideation, experiments, writing, and peer review, but their evaluation is dispersed across tasks, benchmarks, and studies that are difficult to compare directly. We review this literature from the perspective of evaluation design and evidence, covering six targets: literature synthesis, research ideation, executable workflows, scholarly writing and communication, automatic peer review, and end-to-end research. We compare task construction, evidence sources, evaluators, and scoring procedures to explain the capabilities assessed by different designs. Our synthesis highlights three recurring lessons: output checks, process checks, and human studies provide complementary information; evaluator calibration is specific to the property bein
★2Epoch AI 实测前沿模型：真实研究任务仍需人工监督新智元 ASI爆点 · 科研评测 · 原文Epoch AI 用真实研究任务测试 Claude Fable 5.1 和 GPT-6 Astra，发现领先模型仍会在图表风格、参数设置和错误归因等基础环节失误，尚无法独立接管研究工作。
标题：Prior or Feedback? What an LLM Uses When Adapting Neural Operators
来源：arXiv
链接：https://arxiv.org/abs/2610.12325
摘要：Do LLM scientific agents rely only on their initial task context, or do they adapt their decisions in response to experimental feedback? We study this question in neural operator adaptation, where a large language model (LLM) selects fine-tuning configurations under a limited trial budget. Across transfers within and between partial differential equation (PDE) families, the LLM achieves lower held-out test nRMSE than random search and Bayesian optimisation in nearly every matched comparison. Endpoint performance alone cannot distinguish what happens, so we verify each attribution with controlled interventions. Before observing any validation score, the LLM's first configuration already ranks near the top of the corresponding random-search pool, indicating a useful initial bias. A complemen
标题：A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box Optimization
来源：arXiv
链接：https://arxiv.org/abs/2610.12183
摘要：Black-box optimization (BBO) arises in many scientific and engineering problems where objective evaluations are expensive and limited. Recent large language model (LLM) agents offer a new way to approach BBO by combining task semantics, computation, optimization tools, and feedback-driven decision making, showing great potential due to the integration with mathematically rigorous tools. However, existing agentic BBO studies use different task domains and system configurations, making their results difficult to compare and the effects of individual design choices hard to isolate. We therefore introduce AgenticBBO-Bench, a cross-domain benchmark for agentic BBO spanning synthetic functions, hyperparameter optimization, database tuning, chip design, and molecular design under a unified finite
MindFlow 将开放式科研构思显式表示为图结构的“Flow in Mind”，其中节点由模块化思维算子构成，并统一建模为概率思维超网络。给定研究主题后，控制器动态采样不同思维流程，生成兼顾新颖性、合理性与可行性的候选创意。框架采用锦标赛式相对排序处理开放式、多目标评价，使控制器逐步提高高质量构思路径的采样概率。论文还提出联合考察问题发现与问题解决能力的评估协议，避免仅依据标题或摘要判断创意质量。
标题：MindFlow: Mind Supernet Powered Thinking Flows for Research Idea Innovation
来源：arXiv
链接：https://arxiv.org/abs/2610.11966
摘要：Research idea innovation is a fundamental engine of scientific progress, yet it remains difficult to generate and evaluate in a scalable and controllable way. This challenge lies in its inherently open-ended and multi-objective nature, where ideas should balance novelty, plausibility and feasibility. While recent LLM-based approaches have made progress through carefully designed prompts or agent pipelines, they are constrained by predefined, static ideation workflows. To address this limitation, we propose MindFlow, a framework that explicitly formulates ideation as a graph-structured Flow in Mind, which is composed of modular thinking operators and modeled by a probabilistic mind supernet. Given a research topic, a controller dynamically samples thinking flows to generate candidate ideas.
标题：PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives
来源：arXiv
链接：https://arxiv.org/abs/2610.11803
摘要：Self-improving learning systems adapt data selection, optimization, and auxiliary symbolic components, inducing nonstationary objectives outside standard learning assumptions. We introduce \textsc{PRAXIS}, a co-evolutionary framework that models generators, learners, and symbolic archives as interacting dynamical processes. We prove that KL-constrained generator updates and controlled archive-weight movement bound one-step objective drift, that archive updates suppress a program relative to any fixed comparator with a persistent cumulative utility advantage under sub-Gaussian noise, and that stochastic gradient descent achieves an average-stationarity guarantee whose degradation is governed by cumulative objective drift. Experiments across visual robustness, relational graph reasoning, and
标题：EvoAlloc: A Self-Evolving Resource Allocation Agent for Efficient Program Evolution
来源：arXiv
链接：https://arxiv.org/abs/2610.12086
摘要：LLM-based program evolution relies on evaluation feedback to guide the iterative search for high-performing programs. However, evaluation is often computationally expensive, making it essential to allocate limited resources to candidates that can most effectively advance the search. Existing LLM-based methods typically rely on fixed allocation strategies throughout the search, potentially wasting resources on low-value candidates while overlooking promising ones. We propose EvoAlloc, a self-evolving resource-allocation agent that learns from search experience to revise its strategy for allocating computational resources across candidates. EvoAlloc periodically consolidates prior search and allocation outcomes into reusable experience, which informs subsequent strategy revisions. It further
标题：Not Every Change Is Necessary: Recoverable Drift in Large Language Model Unlearning
来源：arXiv
链接：https://arxiv.org/abs/2610.11915
摘要：Machine unlearning in large language models aims to remove unwanted knowledge while preserving the model's remaining capabilities. Although existing methods use retention objectives or restrict where edits occur, achieving the desired forgetting level can still leave collateral changes that impair non-target behavior. Our recovery comparisons suggest that some of these changes can be reversed while preserving observed forgetting performance. In this work, we present Propose-Then-Project Unlearning (PTP-U), a framework that combines targeted forgetting with the recovery of non-target capabilities. PTP-U first applies local analytic edits to weaken target knowledge associations, then aligns non-target output distributions with those of the original model to recover capabilities while maintai
标题：Lamarck's Driving School: Discovering Autonomous Driving Training Strategies through Evolutionary Competition
来源：arXiv
链接：https://arxiv.org/abs/2610.11662
摘要：Autonomous driving capabilities depend strongly on the distribution of scenarios encountered during training. Existing methods commonly construct or dynamically adapt training scenario distributions using surrogate criteria such as realism, difficulty, or risk. However, these predefined surrogates may misrepresent training value, leading to inefficient use of training resources. To address this limitation, we propose a Lamarckian evolutionary framework that replaces surrogate-based guidance with competition among candidate distributions. We formulate training strategy discovery as a multi-stage bilevel optimization problem and use Lamarckian evolution algorithm to approximate its solution. At the outer level, Darwinian crossover, mutation, and selection explore the scenario distribution sp
标题：OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport
来源：arXiv
链接：https://arxiv.org/abs/2610.12375
摘要：Agents are deployed in applications from trip planners and stock trading to IT incident triage. In most cases, LLM agents work autonomously with minimal rule-based safeguarding, leading to cost and safety issues from irreversible actions. Recent works resolve this either by using a safeguard agent to monitor behavior or evaluating logs post-hoc. The first adds cost and latency to every step; the second delivers its verdict after the run, when tokens are burned and damage is done. To overcome this, we propose OnTrack, a streaming monitoring mechanism that compares an agent's steps and dependencies against recorded successful runs to alert users or block the agent in about a millisecond per step. We study this problem in three regimes of decreasing access: full reference access (historical r
标题：Overcoming Prior Barriers: Supervised Fine-Tuning under Long-Tail Distribution
来源：arXiv
链接：https://arxiv.org/abs/2610.12345
摘要：Supervised fine-tuning (SFT) adapts pretrained large language models (LLMs) to downstream tasks, but the required concepts can receive substantially different levels of pretrained support. Frequent concepts are more likely to be well learned, whereas rare concepts may remain weakly represented. We introduce a novel notion named prior barrier to quantify how strongly the pretrained model supports competing concepts over the target concept. We observe that prior barriers follow a long-tail distribution, placing head and tail concepts at different starting points for SFT: head concepts face lower prior barriers, whereas tail concepts require additional instructions to overcome their higher prior barriers. Our theoretical analysis further derives a predictive risk bound for SFT under long-tail
标题：MetaOPD: Meta-Learned Token Weighting for On-Policy Distillation
来源：arXiv
链接：https://arxiv.org/abs/2610.11989
摘要：On-policy distillation (OPD) trains a student on its own generated responses using token-level teacher supervision. However, uniform weighting overlooks differences in token learning value, while existing weighting methods rely on predefined mappings from prediction signals to token weights. These mappings are not learned from the effectiveness of the resulting student updates, limiting their ability to adapt to evolving learning needs. In this paper, we propose MetaOPD, a bilevel optimization framework that jointly learns the student model and a lightweight token-weighting network. The inner objective updates the student through weighted OPD, while the outer objective optimizes the weighting network using validation loss on reference solutions after a virtual student update. Differentiati
标题：DIAL-OPD: Learning More from Fewer Tokens in On-Policy Distillation
来源：arXiv
链接：https://arxiv.org/abs/2610.11659
摘要：On-policy distillation (OPD) supervises student-generated trajectories with token-level teacher signals. Its sampled-token variant avoids the cost of full-vocabulary probabilities. Yet we find that training on fewer tokens can outperform full-token OPD, challenging the intuition that more supervision improves learning. This motivates selecting tokens by learning value. Existing disagreement-based criteria ignore probability scale: tokens assigned negligible probability by both models, termed low-low tokens, can receive large log-ratio rewards and hinder learning. We propose DIAL-OPD, a token-selection method that bridges log-probability and probability spaces by weighting reward magnitude with the logarithmic mean of teacher and student probabilities. A parameter beta controls this weighti
标题：MAP4CS: A Multi-dimensional Data Pruning Framework for Efficient Code Retriever Fine-tuning
来源：arXiv
链接：https://arxiv.org/abs/2610.11727
摘要：Retrieval-Augmented Generation (RAG) has become a cornerstone in software engineering for enhancing Large Language Models (LLMs) with domain-specific knowledge. However, adapting retrievers to evolving code repositories remains challenging due to the noise and redundancy inherent in massive code corpora. Standard fine-tuning on the full corpus is computationally expensive and often leads to sub-optimal performance due to negative transfer from low-quality samples. Conversely, simple random sampling fails to guarantee data representativeness. To address these challenges, we propose MAP4CS (Multi-dimensional Awareness Pruning for Code Search), an adaptive data pruning framework. MAP4CS identifies a small, high-quality core subset by integrating syntactic structure, semantic diversity, and di
标题：Is Real-World Training Data Necessary for Generalist Graph Anomaly Detection?
来源：arXiv
链接：https://arxiv.org/abs/2610.12167
摘要：Generalist graph anomaly detection (GAD) aims to build a foundation model that detects anomalies on arbitrary unseen graphs without retraining or fine-tuning. Sufficient data are essential for foundation model training, yet generalist GAD still faces a data shortage, as real-world anomalous graphs are scarce and costly to collect and annotate. To fill this gap, we propose AG-FORGE, an Anomalous Graph generation Forge for automatic synthesis of anomalous graphs, exploring the feasibility of synthetic data-driven training for generalist GAD. Empirically, we find that synthetic data can achieve performance comparable to real-world training, but fail to push the performance boundary further due to the limited capacity of existing methods. To further unlock model capacity as training data scale
