BEAKER: An Expert-Curated Benchmark for Embodied Brains in Self-Driving Chemical Laboratories
Fei Lin, Tengchao Zhang, Ziyang Gong, Xiaotong Yu, Bohan Zhang, Yifan Zhou, Qihao Yang, Ji Dai 等 18 位作者
AI 速览：作者提出并构建了专家人工标注的 BEAKER 基准，用于评测多模态大模型在自驱动化学实验室中的执行导向具身能力。
Self-Driving Chemical Laboratories (SDCLs) are moving chemical experimentation toward embodied automation, where machines must perceive laboratory scenes, reason about experimental states, and act under physical, procedural, and safety constraints. Although Multimodal Large Language Models (MLLMs) are increasingly considered as embodied brains for such systems, existing laboratory benchmarks mainly focus on safety, anomaly detection, or general scientific understanding, leaving their execution-oriented embodied capabilities underexplored. To fill this gap, we introduce BEAKER, a Benchmark for evaluating Embodied Actionability, Knowledge, Experimentation, and Reasoning in SDCLs. BEAKER contains 1,000 expert-curated question-answer samples across 11 image and video task types, covering key judgments required for chemical laboratory execution, including affordance grounding, trajectory planning, equipment, state, and safety recognition, apparatus setup, operation and step reasoning, and temporal procedure reasoning. All questions and annotations are manually created and reviewed by chemistry and embodied Artificial Intelligence (AI) experts, without AI-assisted generation or annotation. We evaluate 28 general-purpose, science-oriented, chemistry-specialized, and embodied MLLMs in an offline, single-turn setting. Results show that current models can follow coarse experimental procedures in videos, but still struggle to ground actions into operable regions and feasible trajectories, to recognize fine-grained laboratory apparatus, states, and safety norms, and to order steps in mid-length procedures. BEAKER thus reveals a clear gap between coarse procedural understanding and the precise, action-grounded perception required for laboratory execution, providing a diagnostic benchmark for future MLLMs in SDCLs.
Motion-guided Contrastive Regularization For Latent World Models
Yiwen Chen, Qiqi Hou, Jisoo Jeong, Yinhao Zhu, Huaizu Jiang, Amir Said
AI 速览：针对JEPA式世界模型中的静态坍缩问题，提出用光流相似度构造软对比目标的运动引导对比正则化，以提升控制相关表征。
JEPA-style world models learn action-conditioned latent dynamics without reconstructing observations, but distributional regularization alone does not ensure that their representations preserve information needed for control. We identify static collapse, a failure mode in which the encoder preserves episode-specific static content while suppressing manipulation-relevant state changes. Because static content is temporally predictable yet diverse across episodes, this failure can persist despite regularizers such as SIGReg. A controlled study on OGBench-Cube shows that introducing randomized backgrounds substantially degrades planning performance while leaving the underlying manipulation task unchanged. To address this problem, we propose motion-guided contrastive regularization. Our approach compares predicted latent transitions with transitions to candidate observations from the same episode, reducing the usefulness of static background identity for distinguishing action outcomes. Optical-flow similarities provide soft contrastive targets, allowing candidates with similar motion to receive similar supervision. Combined with latent prediction and SIGReg, our objective encourages representations to preserve within-episode state distinctions and similarities between related transitions. Our method improves planning success over LeWorldModel (LeWM) from 52% to 78% on OGBench-Cube with randomized backgrounds and from 14% to 54% on LIBERO-Object, while maintaining comparable performance on four other visual-control benchmarks.
AI 速览：该论文提出首个面向通用多模态大语言模型的分子毒性修复基准任务ToxiMol及自动评估框架ToxiEval，并系统评测43个主流模型。
Toxicity remains a leading cause of early-stage drug development failure. Despite advances in molecular design and property prediction, the task of molecular toxicity repair, generating structurally valid molecular alternatives with reduced toxicity, has not yet been systematically defined or benchmarked. To fill this gap, we introduce ToxiMol, the first benchmark task for general-purpose Multimodal Large Language Models (MLLMs) focused on molecular toxicity repair. We construct a standardized dataset covering 11 primary tasks and 660 representative toxic molecules spanning diverse mechanisms and granularities. We design a prompt annotation pipeline with mechanism-aware and task-adaptive capabilities, informed by expert toxicological knowledge. In parallel, we propose an automated evaluation framework, ToxiEval, which integrates toxicity endpoint prediction, synthetic accessibility, drug-likeness, and structural similarity into a high-throughput evaluation chain for repair success. We systematically assess 43 mainstream general-purpose MLLMs and conduct multiple ablation studies to analyze key issues, including evaluation metrics, candidate diversity, and failure attribution. Experimental results show that although current MLLMs still face significant challenges on this task, they begin to demonstrate promising capabilities in toxicity understanding, semantic constraint adherence, and structure-aware editing.
LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis
Haoxiang Cao, Yuxuan Zhang, Penghui Du, Bo Li, Jiajiong Cao, Yingying Fan, Zimeng Wu, Yifu Luo 等 19 位作者
AI 速览：作者提出 LongBanana，一个由专家设计并验证的长上下文多参考图像合成基准，并配套 LongBanana-Gen 智能体框架来评测和提升这类生成任务。
We introduce LongBanana, to our knowledge the first expert-designed and expert-verified benchmark for long-context multi-reference image synthesis in the age of agent-driven image generation. The task requires retrieving dispersed visual evidence, rejecting distractors, binding entities and relations, and composing one coherent final image from multiple references. LongBanana contains 90 carefully designed tasks organized by a 3-by-3 taxonomy; each task includes about 32 reference images, 20 constraints, and 50 evaluation items, yielding 4,670 expert-specified atomic evaluation items in total. Its graph-grounded evaluation design instantiates each task with reference images and an entity-relation graph, then converts node, edge, and whole-graph obligations into checklist items for Content Fidelity, Relational Binding, and Global Constraint Satisfaction. Reviewed expert labels include a 15% double-labeled subset with 89.4% exact agreement and 0.83 quadratic weighted kappa. We further introduce LongBanana-Gen, an agent-harness framework for long-horizon image generation; experiments show that minimal agent harnesses improve over direct generation, while a ReAct-Plan-Exec-Reflect (RPR) workflow yields harness-dependent gains by strengthening evidence grounding and repair. Together, LongBanana provides an auditable testbed for measuring dense visual evidence use and a shared target for developing stronger agent-driven image-generation systems. Code and data are available at the project page: https://long-banana.com/.
LLMs Don't Know What They Know: Benchmarking Temporal Self-Knowledge for Financial Backtesting
Lingfei Qian, Yan Wang, Xueqing Peng, Yupeng Cao, Yuechen Jiang, Zhuohan Xie, Anke Xu, Wenbo Cao 等 15 位作者
AI 速览：作者提出 FinDendro 基准，检验大语言模型对自身时间性知识边界的自述是否可信，并发现自述普遍低估其实际知识。
Financial backtests of large language models (LLMs) are meaningful only if the model did not already know the outcomes it is asked to predict. Practitioners guard against this by trusting a vendor's stated knowledge cutoff, by asking the model what it knows, or by instructing it to answer as of a past date. To verify whether any of these safeguards can be trusted, we introduce FinDendro, a benchmark of 2,226 dated U.S. facts from 1999 to 2026, covering monthly macroeconomic series, month-end stock prices and Federal Reserve decisions. Each fact yields four prompts: a value question, Recall and Predict questions that pair each candidate value with a claim of knowing or guessing without a declared date, and a Predict question with a historical declared date. Controls for persistence and dominant outcomes keep chance accuracy near 0.27. Results show that for most frontier models, self-reports understate knowledge. GPT-5 answers 76% of controlled facts correctly but claims to know 38%, and answers 61% of the facts it says it does not know. Knowledge claims can persist beyond the measured knowledge boundary, by up to seven quarters, and a declared date removes most of the claims of prior knowledge while leaving accuracy unchanged. Backtests inside a model's training window therefore cannot be certified leak-free by self-report, instruction or stated cutoff. These findings motivate behavioral measurement of temporal knowledge boundaries, and FinDendro provides a benchmark and procedure for selecting backtest periods based on demonstrated factual knowledge.
What Should Data Teach? Moving Bottlenecks Across Circuit, Store, and Use
Yixiao Chen, Ke Cheng, Yue Liu, Jun Zhang, Jie Jiang
AI 速览：作者提出一种“电路视角”的数据诊断原则，将语言模型训练中的瓶颈分为计算形成、内容可用与路径选择三类，并报告相应数据干预在350M模型实验中的效果。
What should data teach a language model at a particular point in training? A circuit view reveals three distinct bottlenecks: forming a computation, making its required content available, and selecting among available routes. A shared diagnosis-to-data principle connects them: localize the missing operation, preserve its causal relation, vary shortcut-bearing context, and re-audit the residual. Formation-sensitive selection and prerequisite ordering accelerate a binding-matching-transport path; a brief early prefix from the same training multiset retains a validation advantage through 100B tokens. Availability counterfactuals then distinguish writing content from invoking available memory, while paired supervision and context-opportunity ranking improve matched route decisions and long-context answer likelihood. A continuous 350M-model experiment connects the three interventions on the same facts: early circuit training improves subsequent learning, and the complete sequence outperforms stage-replacement controls on facts withheld from Use teaching. Independent query surfaces and opposed-source decisions expose conditional arbitration as the remaining frontier. Together, these results show why a change in the limiting operation calls for a change in supervision, not merely a new ranking of difficult examples.
Embodied Internet of Things (IoT) systems are promoting the adoption of task-oriented semantic communication (SemCom), in which the trustworthy task decisions require both semantic information reliability and source legitimacy. However, existing trustworthy designs in SemCom mainly focus on semantic content recovery and protection, with limited attention to source legitimacy. Motivated by this limitation, this paper establishes a new information exploitation paradigm for integrated SemCom-identity protection to address active source impersonation and passive identity inference by treating each SemCom signal as an inherent carrier of both task-relevant semantic information and transmitter physical identity information. We first examine these two information resources and then analyze their integration feasibility at the observation, processing, and decision levels. To exploit these resources for trustworthy decision making, we further develop a dual-branch identity protection mechanism under different operating modes. In the authentication mode, semantic recovery and physical-layer source verification are performed in parallel using the same received observation. In the concealment mode, identity exposure is suppressed while semantic recovery quality is maintained to support anonymous sensing reports. The resulting outputs are jointly interpreted to support task decisions under mode-specific identity protection requirements. Finally, two representative case studies on active source impersonation and passive identity inference are presented to evaluate the proposed framework.
Coding agents powered by models such as GPT-6-Astra connect spatial reasoning with executable robot behavior, raising the question of how such agents can produce lasting improvements in robot capabilities from physical interaction. In this work, we study embodied recursive self-improvement (RSI) through real-world exploration, code refinement, and deployment using PyRUA-REAL, a framework in which coding agents retain interaction experience as persistent memory and reusable executable skills that inform subsequent exploration and improvement. Our empirical study places physical manipulation at the center of the improvement process, examining how agents collaborate, what kinds of execution feedback support effective code refinement, and how task specifications and previously acquired skills influence robustness and task transfer. We further investigate the efficiency and capability of the resulting programs when deployed without involving an online agent for decision-making. Across eight real-world tasks, PyRUA-REAL achieves an average success rate of 91.1%. In complementary evaluation on 42 tasks from the RoboDojo benchmark, it achieves an average success rate of 84.4%. This study provides empirical guidance for turning physical exploration into reusable robot skills that support continued improvement and robust, efficient deployment.
Yequan Bie, Dengyang Jiang, Zhixuan Chen, Xiaozhuan Liang, Haoxuan Che, Quande Liu, Xi Chen, Hao Chen
Generative video models are increasingly used as visual world simulators, yet their primary strength lies in rendering plausible pixels rather than reasoning about how a world should evolve. Coherent world evolution requires more than visual fidelity: when an event occurs, a model must infer its consequences for the entities in the world and then render the resulting state. For example, after a wild beast appears, the model should not only depict the wild beast, but also infer whether a nearby person should flee, take cover, or fight back based on the realized scene. Existing diffusion-based video models lack an explicit mechanism for maintaining world state and reasoning over such event-conditioned consequences, often producing visually plausible but behaviorally static or causally inconsistent outputs. To make this reasoning gap concrete and measurable, we propose a new problem, Out-of-Sight Causal Evolution, which requires a model to render an event, reason about its consequences for the evolving world state, and reveal a causally plausible outcome as the world continues to evolve, even out of sight. As a baseline reference solution, we introduce EvoWorld, a brain-renderer collaboration framework in which a vision-language model serves as the world brain: it observes generated segments, maintains an evolving account of the world, and reasons about plausible state transitions, while a segment-autoregressive video renderer realizes them visually. We further introduce EvoWorld-Data and EvoWorld-Bench for training and evaluating event realization, causal reactions, and revisit consistency. Experiments show that EvoWorld substantially improves event-conditioned causal evolution over recent models while preserving visual quality and consistency, establishing a reference point for reasoning-guided, autonomous world evolution beyond the current field of view.
Generalist robot policies are trained on static offline datasets, yet physical manipulation unfolds sequentially across stages. Because a weakness in a downstream stage remains unobservable until the policy passes earlier ones, each resolved bottleneck exposes the next hidden failure mode. This staged structure calls for Recursive Self-Improvement (RSI), in which a policy iteratively executes, diagnoses failures, and updates itself. Moreover, because task completion behaves like a product of stage-level pass rates, recursive gains can multiply across generations. In physical environments, three bottlenecks obstruct compounding: terminal execution feedback rarely localizes the stage that caused a failure, failed trajectories do not supply the corrective action, and updating shared parameters can degrade previously mastered skills. We propose Compounding Physical RSI, a self-evolving policy framework that turns execution failures into heritable capability. First, automated capability-gap assessment traces failed rollouts to the task stage where progression breaks down and formulates a structured improvement objective. Second, conditioned on this objective, our remedy stage curates validated, stage-relevant experience and refines the policy via behavior-regularized offline reinforcement learning, without requiring costly online rollouts at each optimization step. Third, to safeguard existing skills, our retention stage constructs weight-space candidates between the inherited reference policy and the refined policy, evaluates them across the task suite, and promotes the best candidate as the reference state for the next generation. Evaluated across all 42 tasks of the RoboDojo-Sim benchmark, CSU-AI-07 ranks first on the leaderboard with a three-seed mean Overall Score of 26.15 and Success Rate of 20.70%, outperforming the next-best entry by 5.92 points and 5.82 percentage points.
Human Motion Prediction (HMP) aims to predict future human poses at different moments according to observed past motion sequences. Previous approaches mainly treated the prediction of different temporal moments as a single prediction task and learned the predictions of varied moments simultaneously, which would encounter a main limitation: the learning of short-term predictions (referring to ``near-future'' prediction) could be hindered by the predictions of long-term (referring to ``far-future'' prediction) motions. In this paper, we develop a novel temporal continual learning framework called Continual Prior Compensation (CPC) to progressively train HMP models, in which we divide the prediction task of motions corresponding to varied temporal moments into several subtasks and train the model in a multi-stage manner. To mitigate the prior information forgetting in the progressive training, we further introduce a learnable random variable Prior Compensation Factor (PCF) to explicitly measure the prior knowledge loss. We theoretically show that the PCF can be efficiently learned together with the model parameters by minimizing a reasonable upper bound of the objective function. The proposed CPC is further enhanced to estimate the prior information loss for each subtask and a new framework called Continual Prior Compensation++ (CPC++) with Fine-Grained Prior Compensation Factor (FGPCF) is finally developed. Our CPC and CPC++ frameworks are quite flexible and can be easily integrated with different HMP backbone models and adapted to various datasets and applications. Extensive experiments on three HMP benchmark datasets using multiple SOTA HMP backbones (PGBIG, siMLPe, MotionMixer, and LTD) demonstrate the effectiveness and flexibility of our frameworks.
Continual agents must handle urgent interaction while retaining alternatives, replaying errors, and committing safe long-term changes. A previous digital-life framework distinguishes online (busy) and offline (idle) channels, but leaves open whether these are separate programs or phases of a common process and how their operations should be ordered. Motivated by recent observations of rapidly alternating perceptual representations in primate cortex, we develop a temporal organization layer with phase-tagged micro-epochs, an observable duty cycle, independent content and resource controls, and three nested clocks separating candidate exploration from durable commitment. We give a projected duty-cycle update, state its deterministic descent guarantee and the additional noise term required for a stochastic guarantee, and specify coupled busy/idle execution contracts. Five synthetic experiments (64 seeds each) examine descent and noise, objective misalignment, gain asymmetry, resource coupling, and dropped replay logs. In the gain/noise model the best tested gain ratio is 1.55; decoupled scheduling yields 26.6% versus 11.1% success on a constructed urgent/high-load subset. A frozen-backbone Qwen3-8B prototype further checks orchestration, but its proxy outcomes are not task-benchmark evidence. The results establish a testable scheduling design and identify its conditions of validity, not machine consciousness or general-purpose agent superiority.
Jinjian Jiang, Jing Peng, Rehan Deen, Yu Cao, Ziquan Liu, Muhammad Salman Haleem, David Mguni
When an object category is well represented in one domain but scarcely observed in another, adapting a diffusion model to the target can be unreliable. Yet the same source-target relationship may have been observed for other categories. We ask whether this recurring relation can be learned and reused for a new semantic task. We introduce Meta-Diff, which embeds source and target distributions in a shared low-dimensional task space, learns an affine transport map across paired training tasks, and generates from the transported source coordinate using a shared diffusion backbone. At deployment, it requires source examples and the identity of a previously learned relation, but neither target images nor target-specific optimisation. Gaussian-mixture experiments examine how transfer depends on task information retained by the encoder, relation predictability and the size of the source--target shift. On held-out CIFAR-100 validation tasks, correct-source transport achieves 13.9% lower mean sliced Wasserstein distance than wrong-source transport. In the one-target final-test comparison, it achieves 14.7% lower mean distance than target-only adaptation and 32.7% lower than full fine-tuning under validation-selected settings. These gains have clear limits: directly applying the known corruption to clean-source generations performs better on CIFAR, and requested-class identity remains weak. Office-Home shows domain alignment without reliable object identity; clinical stress tests show no resolved source-specific advantage. The results characterise when reusable task-space relations help and where their generative limitations remain.
Long-term time-series forecasting supports decisions in energy, transportation, and other applications where reliable predictions are needed beyond the observed data. Although forecasting research has advanced model architectures, fixed activation configurations provide limited guidance on how activation choices should vary across datasets and prediction horizons. This study aims to establish when task-specific activation choices improve forecasting accuracy and how previous experiments can guide the selection of suitable configurations for new tasks. We develop a calibrated selection policy that learns a reliable default from prior experiments, adapts the activation function and its location to each task, and retains the default when evidence for a change is insufficient. In an evaluation supported by 1,225 Informer training runs across 17 dataset variants and 51 forecasting tasks, the policy reduces mean squared error by 7.86% on average relative to the conventional fixed-activation baseline and improves 72.55% of tasks. These findings establish activation choice and placement as practical opportunities for improving existing forecasting models and show how accumulated experimental evidence can guide configuration decisions without repeating an exhaustive search for every task.
Ximo Zhu, Ruiqi Liu, Yi Han, Zhengan Wu, Gengsheng Li, ZiAn Wang, Zhu Yuantian, Jin Jiang 等 9 位作者
Proprietary language models exhibit strong reasoning and coding capabilities, but their output distributions are inaccessible, making fine-grained distillation methods that rely on teacher logits difficult to apply directly to black-box models. We observe that rich textual feedback from a black-box teacher can be treated as privileged information (PI) and converted, through a PI-conditioned privileged teacher, into dense supervision signals on the student's own trajectories. However, what PI should be acquired from the teacher to most effectively support student learning remains unclear. PI strategies that differ in where they focus, what information they request, and how much detail they provide can have substantially different effects on the current student, and their relative effectiveness can further change as training progresses. Based on this observation, we propose Evolving Privileged Information Strategies for Black-Box On-Policy Distillation (EvoOPD), which treats PI acquisition itself as an optimization problem and maintains a PI Strategy Bank that evolves with the student state. EvoOPD uses the measured repair utility of PI on the current student to adapt the focus, information type, and granularity of teacher feedback, allowing the acquired information to continually match the student's learning needs. We systematically evaluate EvoOPD on mathematical reasoning and code generation with Qwen3-1.7B, 4B, and 8B students. EvoOPD achieves the best overall performance in all six model-task settings, improving over the strongest baseline in each setting by an average of 2.14 percentage points in mathematics and 1.30 points in code. Further analysis shows that the relative effectiveness of PI acquisition strategies changes with student state, making a fixed strategy difficult to sustain as the best choice throughout distillation.
Eigenvector centrality evaluates the importance of vertices through adjacency relations, while higher-order subgraph structures provide additional information for centrality analysis. This motivates the simultaneous evaluation of vertices and subgraphs within a unified spectral framework. In this paper, we propose vertex-subgraph eigenvector centrality (VSC), which jointly determines the centralities of vertices and connected induced subgraphs of order three. The vertex centrality depends on the centralities of the subgraphs containing it and those of the other vertices in these subgraphs, whereas the subgraph centrality depends on the centralities of its constituent vertices. To distinguish induced paths P3 from triangles K3, the capacity of each subgraph is defined as its number of edges. The resulting coupled nonlinear eigenvalue equations are represented by a fourth-order symmetric nonnegative tensor. We prove that this tensor is weakly irreducible for every connected graph with at least three vertices, which guarantees the existence and uniqueness, up to scaling, of a positive Perron eigenvector. Numerical experiments on four real-world networks compare VSC with classical centrality measures and show that capacity weighting has a greater effect on subgraph rankings than on vertex rankings.
Group Relative Policy Optimization (GRPO) has advanced generative model post-training in verifiable domains such as code and mathematics, but evaluating video quality, especially emotional expression, often depends on subjective aesthetics and has no uniquely correct target. Existing approaches typically rely on fixed rewards for visual quality, motion quality, and text alignment that poorly capture emotional intent and can invite reward hacking, while RLHF is difficult to scale and subject to individual aesthetic differences that complicate the calibration of facial, body motion, and chromatic cues across narrative contexts. We introduce EmoReward, a new self-evolving training mechanism that distills evidence reliability from reward traces collected during an actual training run to adaptively guide multi-reward weight allocation in subsequent training by determining which emotional evidence to trust and how much to weight it. EmoReward first plans specific grounding cues from the emotional prompt to guide evaluation, then uses heterogeneous signal primitives to score the generated video during the act stage. Reflective Reward Gating calibrates their contributions through instance-level gates for phase, face confidence, conflict, and subtlety, then renormalizes weights over the active set to produce the sample's fused reward. RewardMemory then estimates each primitive's reliability from rollout scores and outcomes within each emotional regime in the finished training run, then fixes bounded multipliers that calibrate fusion weights in the next training run without updating reward models. Experiments on HunyuanVideo 1.5 show improvements across multiple RL algorithms and compare EmoReward with other agentic reward methods under SAGE-GRPO. Ablations and transfer experiments on SD 3.5 and FLUX further support adaptive reward calibration for visual generation.
Time-series forecasting is foundational to a wide spectrum of web-scale services, spanning personalized recommendations, energy scheduling, and infrastructure capacity planning. Accurate long-term time series forecasting (LTSF) requires models to capture long-range temporal dependencies while adapting to periodic patterns that vary across variables. Although frequency-aware methods offer a global view of temporal dynamics, they typically rely on inflexible heuristic designs, such as predefined period lengths, manually selected or randomly sampled top-k frequency components, and separate spectral transformations. Consequently, they fail to account for time-frequency heterogeneity, where variables that appear synchronized in the time domain can differ substantially in the frequency domain. Meanwhile, Mamba-based forecasters, despite their efficiency in long-range modeling, introduce frequency information only as auxiliary signals, leaving spectral evidence disconnected from state transitions. In this work, we propose AdaMamba, a novel approach that explicitly internalizes input-dependent frequency information into the modeling of temporal dependencies. AdaMamba comprises an interaction learning module and an adaptive frequency-gated state space module: the former captures cross-variable dependencies and multi-scale temporal context, while the latter generates context-dependent frequency bases and unifies temporal forgetting into a single time-frequency gate that directly modulates state evolution. Extensive experiments on seven public LTSF benchmarks and four domain-specific datasets demonstrate that AdaMamba consistently outperforms state-of-the-art baselines while maintaining competitive computational efficiency.
Skills let language agents reuse experience across tasks as natural-language documents. Recent work evolves skills from interaction, but existing methods have limitations when skills evolve online. Prior methods update a skill offline from a batch of trajectories collected over many tasks. Some then adopt the new skill without any check, which can degrade the skill over time. Others keep it only if it improves performance on a held-out validation set, which is too slow and costly to run after every update. We introduce EvoGuard, a pairwise skill selector that mitigates skill degradation at low latency and cost. We first run the agent iteratively to build pools of skills and score each skill on a private test set to obtain utility labels. We pair the skills and use rejection sampling to keep only pairs with a clear utility gap. The selector is trained on the logits of two label tokens to pick the better skill. At deployment it decides in a single forward pass, which is significantly faster than evaluation on the private test set. Compared with adopting every new skill, EvoGuard improves success by 11.0% on ALFWorld, 5.9% on ScienceWorld, and 2.7% on LiveCodeBench.
The rapid proliferation of LLM-based autonomous agents in real operating system environments introduces a new category of safety risk beyond content safety: behavior jailbreak, where an adversary induces an agent to execute dangerous OS-level operations with irreversible consequences. Existing benchmarks either evaluate safety at the semantic layer alone, missing physical-layer harms, or fail to isolate test cases, letting earlier runs contaminate later ones. We present LITMUS (LLM-agents In-OS Testing for Measuring Unsafe Subversion), a benchmark addressing both gaps via a semantic-physical dual verification mechanism and OS-level state rollback. LITMUS comprises 819 high-risk test cases organized into one harmful seed subset and six attack-extended subsets covering three adversarial paradigms (jailbreak speaking, skill injection, and entity wrapping), plus a fully automated multi-agent evaluation framework judging behavior at both conversational and OS-level physical layers. Evaluation across frontier agents reveals three findings: (1) current agents lack effective safety awareness, with strong models (e.g., Claude Sonnet 4.6) still executing 40.64% of high-risk operations; (2) agents exhibit pervasive Execution Hallucination (EH), verbally refusing a request while the dangerous operation has already completed at the system level, invisible to every prior semantic-only framework; and (3) skill injection and entity wrapping attacks achieve high success rates, exposing pronounced agent vulnerabilities. LITMUS provides the first standardized platform for reproducible, physically grounded behavioral safety evaluation of LLM agents in real OS environments.
Large language models have made rapid progress in generating machine-checked proofs for well-defined mathematical problems. Extending this progress to research mathematics requires more than solving isolated theorems. Research involves choosing questions, developing definitions and conjectures, connecting results, and combining formal proofs with expert judgment. This survey examines how LLM-driven systems can support these broader activities. We review formal-mathematics tasks, datasets, and autoformalization, and organize proof-generation methods into three overlapping areas: training strategies, inference-time search and adaptation, and agentic workflows. We then examine benchmark results, selected research projects, and platforms for collaborative mathematical work. A central distinction runs throughout the survey: checking a formal proof does not by itself establish that it captures the intended problem or makes a new mathematical contribution. We therefore distinguish theorem benchmarks, repository tasks, formalization projects, and research claims, and discuss the evidence needed to assess each. Finally, we identify five priorities for progress: better data and evaluation, representations of mathematical relationships, conjecture and concept generation, reliable tool integration, and human--AI collaboration. Together, these perspectives connect advances in proof generation to the practical demands of mathematical research.
Xingchen Zou, Maonan Wang, Kemou Jiang, Jie Li, Yiqi Chen, Xi Chen, Hai Yang, Yuxuan Liang
Existing traffic signal control methods typically rely on structured traffic states produced by dedicated perception pipelines. Emerging visual approaches often use idealized simulator images and convert them into intermediate representations before control, limiting their ability to directly exploit roadside visual observations. To bridge both the visual-domain and perception-to-decision gaps, we introduce SynTra-V and Traffic-VL. SynTra-V converts physics-based traffic simulations into photorealistic multi-view surveillance imagery across diverse road topologies and sensing conditions, and automatic semantic auditing checks that control-critical geometry, vehicle movements, and lane semantics match the simulator state. Traffic-VL is an eight-billion-parameter vision-language model that directly maps synchronized multi-view images and natural-language phase descriptions to signal decisions without intermediate structured states. Building on Traffic-R1's two-stage reinforced reasoning framework, Traffic-VL adapts its vision encoder via offline reinforcement learning on SynTra-V and freezes it during online visual stepwise trajectory policy optimization in closed-loop simulation. Experiments show that Traffic-VL matches structured-state Traffic-R1 while outperforming image-based controllers and general-purpose vision-language models, including Gemini 3 Pro. Evaluations on unseen topologies, degraded sensing conditions, real surveillance footage, and a three-week deployment across ten intersections further demonstrate robust visual generalization and effective constrained online control.
Shaopeng Liang, Biao Zhang, Hongzhen Ye, Zeyang Hou, Ma Jiang, Jintao Tu, yunxi. hz, Zhijia Fang 等 11 位作者
Generative retrieval predicts semantic identifiers for candidate entities. In local retail and food delivery, candidates must also be serviceable for a particular user at a particular time. Location, weather, and traffic reshape merchant and product supply, so a catalog-valid identifier may be infeasible for the current request. Post-generation filtering removes such candidates only after beam capacity and inference computation have been spent. We study retrieval over an explicitly materialized, request-time supply snapshot, where short output sequences leave little opportunity to amortize online constraint preparation. Our RTP-LLM pipeline compiles eligible residual-quantization identifiers into a compressed sparse row prefix index, overlaps CPU compilation with model prefill, and shares the uploaded index across beams. Legal expansions retain original-model log probabilities, and constraint states follow selected beam parents. We formalize snapshot-relative identifier validity and distinguish it from entity-level eligibility and freshness. The implementation generates three-token merchant identifiers. Historical evidence covers daily catalog constraints and subsequent request-scoped serving evaluations. In one matched-beam daily-constraint case, filtered merchants decrease from 170 of 299 candidates to 29 of 299; residual distance filtering motivates request-specific constraints. A different-host comparison with 30,000 supplied entries reports a 0.6 ms service-side latency increment, while client measurements expose additional request costs. Captured requests from three geographic settings characterize behavior across supply sizes and output widths. These observations span development stages and measurement boundaries. They establish a practical implementation and identify bottlenecks, but do not establish compute savings at matched retrieval quality.
Ziyi Jin, Kangjin Yu, Beichen Wang, Can Jiang, Zhongpu Xia, Jie Xu
Feed-forward geometric foundation models enable dense monocular reconstruction without requiring camera calibration, but independently reconstructed submaps can accumulate scale drift, and even a valid revisit can yield an inconsistent cross-submap measurement. We extend VGGT-SLAM 2.0 with fixed-root scale anchoring and short-sequence-assisted direct loop connections. SALAD selects an image pair between the initial submap and each eligible new submap; joint VGGT inference then uses learned implicit semantic-geometric cues to supply a scale-only anchor while preserving the original overlap chain. For loop closure, neighboring image pairs support a consistency check and translation-scale alignment. Each accepted candidate contributes one robust Sim(3) factor between existing center-frame nodes, without an auxiliary loop submap. A 17-sequence comparison lowers mean Sim(3) ATE on nine TUM sequences from 0.1018 to 0.0451 m relative to the VGGT-SLAM 2.0 baseline. Anchoring improves all seven no-loop module ablations. On three Clio scenes, a fixed-loop comparison favors the direct connection pipeline over the tested auxiliary-submap pipeline. A separate candidate audit exposes additional rejections of visually overlapping pairs by the short-sequence check. These comparisons distinguish the gains from long-range scale anchoring and loop connection choices from the limitations of candidate verification.
We identify a fundamental mismatch in empathetic reinforcement learning: support priorities evolve with the dialogue state, yet existing methods typically optimize predefined reward specifications that remain fixed across turns. To model these evolving support priorities, we organize empathetic support along cognitive, affective, and proactive empathy, and propose Context-Adaptive Rubric Evolution (CARE). At each turn, CARE generates a context-adaptive rubric by adjusting both the weights of these three empathy dimensions and their fine-grained evaluation criteria. The rubric generator is trained with turn-level rubric supervision and human preference data through supervised fine-tuning followed by preference-based reinforcement learning, and then serves as an adaptive reward interface for online empathetic RL. Integrated with both RLVER and MICA, CARE achieves state-of-the-art performance across SentientBench, EQBench3, and EMPA under three independent LLM judges. Notably, on EMPA, CARE improves EPM-Idx over the strongest baseline by at least 13 points under all three judges, including an increase from 28.11 to 83.54 under Gemini-2.5-Pro. Further analyses show that learned rubric priorities systematically vary across dialogue stages and user emotions, demonstrating that CARE adapts what is rewarded as support needs evolve. Code is available at https://asa13opp.github.io/CARE-project-page.
Wei-Chieh Huang, Weizhi Zhang, Yuchen Wu, Yankai Chen, Eric Hanchen Jiang, Wooseong Yang, Yiwei Yang, Henry Peng Zou 等 15 位作者
Memory is becoming core infrastructure for long-horizon LLM agents, yet existing evaluations offer limited guidance on which memory substrate, namely the underlying medium in which memory is represented and stored, should be used under different operating regimes. We introduce a unified suite of 26 performance and efficiency metrics, including auxiliary-LLM costs attributed to the write, read, and management stages, and apply it identically to 11 memory methods in a controlled harness that varies only the memory substrate. The methods cover dense and sparse indices, text records, structural stores, hierarchical stores, refinement-based memories, parametric updates, and activation-compatible context mechanisms, and we evaluate them with three backbone models on four benchmark suites spanning user-centric question answering and agent-centric decision-making. Our results show that no single substrate consistently dominates: broad retrieval benefits long-context factual QA, while excessive retrieval can harm sequential decision-making by shifting attention away from action-critical context. Scalability introduces a further routing axis, as substrates that perform well at moderate history lengths can become costly or brittle at longer horizons. These regime-dependent reversals make substrate selection a design axis. Choosing a substrate per observable operating regime improves over the best fixed substrate on our benchmarks, while query-level routing leaves most of the oracle headroom open. The findings provide empirical guidance for designing efficient, reliable, and regime-aware long-term memory for LLM agents.
Zhengyu Hu, Jianxun Lian, Zheyuan Xiao, Max Xiong, Tianfu Wang, Teng Xiao, Yuxuan Lei, Fengqing Jiang 等 13 位作者
Recent advances in large language models (LLMs) have enabled large-scale, high-fidelity social simulations, creating new opportunities for computational social science. However, constructing persona sets that faithfully reflect real-world population diversity remains a key challenge. Existing studies often emphasize agentic frameworks and simulation environments, while paying less attention to persona generation and the biases introduced by unrepresentative persona sets. In this paper, we propose a systematic framework for synthesizing high-quality, population-aligned persona sets for LLM-driven social simulation. Our approach begins by leveraging LLMs to generate narrative personas from long-term social media data, followed by rigorous quality assessment to filter out low-fidelity profiles. We then apply importance sampling to achieve global alignment with reference psychometric distributions, such as the Big Five personality traits. To address the needs of specific simulation contexts, we further introduce a task-specific module that adapts the globally aligned persona set to targeted subpopulations. Extensive experiments demonstrate that our method significantly reduces population-level bias and enables accurate, flexible social simulation for a wide range of research and policy applications.
With the rapid development of autonomous driving systems (ADSes), real-world roads are evolving into mixed-traffic environments where autonomous vehicles (AVs) and other traffic participants (e.g., human-driven vehicles (HDVs), cyclists, and pedestrians) coexist and interact with each other. However, current ADSes still incur various safety violations in mixed traffic scenarios, raising a strong demand for runtime safety assurance frameworks that can continuously monitor hazards and prevent potential violations. Existing ego-centric assurance approaches are insufficient for mixed traffic, where violations often arise from locally reasonable yet jointly incompatible decisions among multiple vehicles and cannot be reliably resolved through independent takeover. To bridge this gap, we propose CoSafe, a modular and ADS-agnostic cooperative runtime safety assurance framework for mixed-traffic autonomous driving to extend runtime safety assurance from ego-centric takeover to scene-level cooperative intervention. CoSafe leverages vehicle-to-everything (V2X) communication to enable roadside units (RSUs) to monitor scene-level hazards and jointly mitigate hazards involving multiple vehicles. We integrate CoSafe with seven state-of-the-art ADSes, and our evaluation on the InterDrive benchmark demonstrates that CoSafe improves the driving score of AVs by 70.03% on average, reduces safety violations by 87.80%, and accurately identifies unsafe AVs with an F3 score of 0.930, while satisfying real-time constraints.
Deep learning-based super-resolution (SR) can reconstruct fine-scale Earth-observation fields from coarse observations, but reconstruction accuracy alone does not provide a calibrated rule for identifying reliable spatial regions. Conformal Mask Calibration (CMC), built on conformal risk control, converts a spatial uncertainty score into a trust mask and controls the expected worst-case local distortion under exchangeability. However, its fixed threshold can become miscalibrated in Earth-observation streams affected by seasonality and episodic events. We propose $\textit{CARM}$ (Conformal Adaptive Risk Masking), a statistical framework for adaptive spatial risk control under temporal distribution shift. CARM maintains experts with heterogeneous update rates and aggregates their threshold proposals online using an asymmetric surrogate loss, balancing rapid adaptation to distribution changes against stability during relatively stationary periods. All updates are causal and use only mature feedback available after a fixed delay, allowing CARM to operate with any SR model that provides a spatial uncertainty score. We prove that the time-averaged mask-conditioned risk converges to at most $α$ at an $O(1/T)$ rate, without distributional assumptions and for any choice of aggregation weights across experts. On a daily South China Sea chlorophyll-$\textit{a}$ SR stream, CARM reduces the empirical violation rate from $43.49\%$ to $9.38\%$ and the moving-average exceedance frequency from $47.29\%$ to $1.22\%$ at $α=0.10$, while retaining a non-trivial trusted region. Experiments on ERA5 wind and OSTIA sea-surface temperature further show consistent relative improvements across multiple SR backbones.
Xi Xiao, Yunbei Zhang, Lin Zhao, Janet Wang, Xinru Jiang, Yanshu Li, Chenrui Ma, Pan Wang 等 10 位作者
Segment Anything Model 3 (SAM 3) can segment objects from text prompts and track them across video, but its memory and computation costs make deployment difficult. Post-training quantization (PTQ) reduces these costs by using fewer bits. However, a quantized model can select the wrong object or lose track of its target across frames. We call this prompt-alignment drift. Controlled tests show that perturbing low-variance feature directions strongly affects target selection. We propose Semantics-Preserving Quantization (SPQ), which gives more calibration weight to concept-sensitive parameters, preserves residual features, and corrects the mean error in video memory. Fixed text concepts guide calibration without task annotations. All three components operate offline and add no inference operators. With a W4A8 image encoder, SPQ improves RefCOCO Pr@0.5 from 42.1 to 72.5 over PTQ4SAM while retaining 54.8 COCO mask AP. Experiments on LVIS, RefCOCO variants, and SA-V show consistent gains. A CLIP study also shows improved image-text retrieval.