Retrieval-augmented generation (RAG) systems require large language models (LLMs) to reconcile discrepancies between their parametric memory—knowledge encoded during training—and contextual inputs provided at inference. When these sources conflict, models often exhibit unstable reasoning and inconsistent factual behavior. We investigate how LLMs resolve such conflicts when the discrepancy arises from temporal misalignment—facts that have changed since the model’s knowledge cutoff—and whether mutability, the changeability of facts, can serve as a mediating signal in this process. To do so, we provide WIKIRECENTCHANGES, a temporally grounded benchmark with stable and recently updated facts derived from Wikidata.Our results show that while models spontaneously produce temporal reasoning for facts that actually changed — but almost never for stable ones — this differentiation rarely propagates to their final predictions. Explicitly prompting them to consider mutability increases references to temporal change but does not improve factual accuracy, revealing a disconnect between verbalized reasoning and prediction behavior. We further show that the failure point is scale-dependent: smaller models rarely detect the underlying conflict, while larger models detect it but fail to act on their mutability judgments.
论文检索
输入标题、作者或关键词，从 9,460 篇学术成果中精准定位
Solver-Independent Automated Problem Formulation via LLMs for High-Cost Simulation-Driven Design
PDF ↗In the high-cost simulation-driven design domain, translating ambiguous design requirements into a mathematical optimization formulation is a bottleneck for optimizing product performance. This process is time-consuming and heavily reliant on expert knowledge. While large language models (LLMs) offer potential for automating this task, existing approaches either suffer from poor formalization that fails to accurately align with the design intent or rely on solver feedback for data filtering, which is unavailable due to the high simulation costs. To address this challenge, we propose automated problem formulation (APF), a solver-independent framework that utilizes LLMs to convert engineers’ natural language requirements into executable optimization models. The core of this framework is an innovative pipeline for automatically generating high-quality data, which overcomes the difficulty of constructing suitable fine-tuning datasets in the absence of high-cost solver feedback with the help of data generation and test instance annotation. The generated high-quality dataset is used to perform supervised fine-tuning on LLMs, significantly enhancing their ability to generate accurate and executable optimization problem formulations. Experimental results on antenna design demonstrate that APF significantly outperforms the existing methods in both the accuracy of requirement formalization and the quality of resulting radiation efficiency curves in meeting the design goals.
We present **AMO-Bench**, an **A**dvanced **M**athematical reasoning benchmark with **O**lympiad level or even higher difficulty, comprising 50 human-crafted problems. Existing benchmarks have widely leveraged high school math competitions for evaluating mathematical reasoning capabilities of large language models (LLMs). However, many existing math competitions are becoming less effective for assessing top-tier LLMs due to performance saturation (e.g., AIME24/25). To address this, AMO-Bench introduces more rigorous challenges by ensuring all 50 problems are (1) cross-validated by experts to meet at least the International Mathematical Olympiad (IMO) difficulty standards, and (2) entirely original problems to prevent potential performance leakages from data memorization. Experimental results across 36 LLMs on AMO-Bench highlights three key findings: (1) high-level mathematical reasoning remains challenging for current LLMs, with even the best-performing model achieving only 63.1% accuracy and most LLMs scoring below 50%; (2) scaling test-time compute remains a highly effective strategy for substantially improving reasoning performances, and (3) open-source models are progressively narrowing the performance gap with proprietary models. Additionally, we conduct further analysis about reasoning efficiency, volatility, and cross-lingual robustness, providing deeper insights behind the reasoning performances.
Large language model agents increasingly rely on memory to support long-horizon interaction, yet existing frameworks expose only a small set of low-level primitives and lack a formal, executable specification for memory control. As a result, higher-order operations such as promotion, consolidation, or lifecycle governance are missing or inconsistently implemented, leading to unpredictable behavior across systems. We introduce Text2Mem, a unified memory operation language that standardizes the translation of natural-language instructions into reliable execution. Text2Mem defines a compact and expressive operation set spanning encoding, storage, and retrieval, and represents each instruction as a schema-based contract with explicit fields and semantic invariants. Validated schemas are parsed into typed operation objects and executed through a unified pipeline that supports both a SQL reference backend and real memory frameworks, enabling safe, deterministic, and portable behavior across heterogeneous systems. We further outline the Text2Mem Benchmark, which decouples schema generation from backend execution to systematically evaluate planning accuracy and execution fidelity. Together, Text2Mem and its benchmark establish a standardized foundation for controllable and reproducible memory management in LLM-based agents.
MulDimIF: A Multi-Dimensional Constraint Framework for Evaluating and Improving Instruction Following in Large Language Models
PDF ↗Instruction following refers to the ability of large language models (LLMs) to generate outputs that satisfy all specified constraints. Existing research has primarily focused on constraint categories, offering limited evaluation dimensions and little guidance for improving instruction-following abilities. To address this gap, we introduce MulDimIF, a multi-dimensional constraint framework encompassing three constraint patterns, four constraint categories, and four difficulty levels. Based on this framework, we design a controllable instruction generation pipeline. Through constraint expansion, conflict detection, and instruction rewriting, we construct 9,106 code-verifiable samples. We evaluate 18 LLMs from six model families and find marked performance differences across constraint settings. For instance, average accuracy decreases from 80.82% at Level I to 36.76% at Level IV. Moreover, training with data generated by our framework significantly improves instruction following without compromising general performance. In-depth analysis indicates that these gains stem largely from parameter updates in attention modules, which strengthen constraint recognition and adherence. Code and data are available in https://github.com/Junjie-Ye/MulDimIF.
Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterized by long interaction traces, inter-agent dependencies, and delayed error manifestation. Existing diagnostic approaches often rely on expensive expert annotation or ”LLM-as-a-judge” paradigms, which struggle to pinpoint decisive error steps within extended contexts. In this paper, we introduce ErrorProbe, a self-improving framework for semantic failure attribution that identifies responsible agents and the originating error step. The framework operates via a three-stage pipeline: (1) operationalizing the MAS failure taxonomy to detect local anomalies, (2) performing symptom-driven backward tracing to prune irrelevant context, and (3) employing a specialized multi-agent team (Strategist, Investigator, Arbiter) to validate error hypotheses through tool-grounded execution. Crucially, ErrorProbe maintains a verified episodic memory that updates only when error patterns are confirmed by executable evidence, without the need for annotation. Experiments across the TracerTraj and Who When benchmarks demonstrate that ErrorProbe significantly outperforms baselines, particularly in step-level localization, while the verified memory enables robust cross-domain transfer without retraining.
Large Language Models (LLMs) exhibit strong general reasoning but struggle in molecular science due to the lack of explicit priors required for precise chemical reasoning. Current solutions inject priors into parameters, but this static coupling hinders rapid knowledge updates and often compromises the model’s general capabilities. To address this, we introduce REAP, a training-free framework that equips LLMs with an external knowledge base, enabling them to reason over retrieved chemical priors dynamically. REAP implements a structured reasoning pipeline that autonomously selects relevant priors from our constructed atom-level knowledge base, retrieves analogue exemplars, and synthesizes these information to guide the LLM’s decision-making. This architecture ensures interpretability and adaptability while preserving the LLM’s intrinsic general intelligence. Experiments show that REAP outperforms current reasoning methods and rivals state-of-the-art training-based models, demonstrating the effectiveness of our framework.
MTMCS-Bench: Evaluating Contextual Safety of Multimodal Large Language Models in Multi-Turn Dialogues
PDF ↗Multimodal large language models (MLLMs) are increasingly deployed as assistants that interact through text and images, making it crucial to evaluate contextual safety when risk depends on both the visual scene and the evolving dialogue. Existing contextual safety benchmarks are mostly single-turn and often miss how malicious intent can emerge gradually or how the same scene can support both benign and exploitative goals. We introduce the Multi-Turn Multimodal Contextual Safety Benchmark (MTMCS-Bench), a benchmark of realistic images and multi-turn conversations that evaluates contextual safety in MLLMs under two complementary settings, escalation-based risk and context-switch risk. MTMCS-Bench offers paired safe and unsafe dialogues with structured evaluation. It contains over 30 thousand multimodal (image+text) and unimodal (text-only) samples, with metrics that separately measure contextual intent recognition, safety-awareness on unsafe cases, and helpfulness on benign ones. Across eight open-source and seven proprietary MLLMs, we observe persistent trade-offs between contextual safety and utility, with models tending to either miss gradual risks or over-refuse benign dialogues. Finally, we evaluate five current guardrails and find that they mitigate some failures but do not fully resolve multi-turn contextual risks.
Agentic Reinforcement Learning (Agentic RL) has achieved notable success in enabling agents to perform complex reasoning and tool use. However, most methods still relies on sparse outcome-based reward for training. Such feedback fails to differentiate intermediate reasoning quality, leading to suboptimal training results. In this paper, we introduce Agent Reasoning Reward Model (Agent-RRM), a multi-faceted reward model that produces structured feedback for agentic trajectories, including (1) an explicit reasoning trace , (2) a focused critique that provides refinement guidance by highlighting reasoning flaws, and (3) an overall score that evaluates process performance. Leveraging these signals, we systematically investigate three integration strategies: Reagent-C (text-augmented refinement), Reagent-R (reward-augmented guidance), and Reagent-U (unified feedback integration). Extensive evaluations across 12 diverse benchmarks demonstrate that Reagent-U yields substantial performance leaps, achieving 43.7% on GAIA and 46.2% on WebWalkerQA, validating the effectiveness of our reasoning reward model and training schemes. Code, models, and datasets will be released to facilitate future research.
Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models
PDF ↗Large Reasoning Models (LRMs) demonstrate strong performance on complex tasks but often suffer from excessive verbosity, known as "overthinking." Existing solutions via reinforcement learning (RL) typically penalize generated tokens to promote conciseness. However, these methods encounter two challenges: responses with fewer tokens do not always correspond to fewer reasoning steps, and models may develop hacking behavior in later stages of training by discarding reasoning steps to minimize token usage. In this work, we introduce Step Pruner (SP), an RL framework that steers LRMs toward more efficient reasoning by favoring compact reasoning steps. Our step-aware reward function prioritizes correctness while imposing penalties for redundant steps, and withholds rewards for incorrect responses to prevent the reinforcement of erroneous reasoning. Moreover, we propose a dynamic stopping mechanism to prevent hacking behavior caused by step merging. Extensive experiments across four reasoning benchmarks demonstrate that SP achieves state-of-the-art accuracy while significantly reducing response length. For instance, on AIME24, SP reduces token usage by 69.7%.
Large language models (LLMs) have driven major advances across domains, yet their massive size hinders deployment in resource-constrained settings. Low-rank factorization addresses this challenge by compressing models to effectively reduce their computation and memory consumption while maintaining accuracy. While these compressed models boast benign performance and system-level advantages, their trustworthiness implications remain poorly understood. In this paper, we present the first comprehensive study of how low-rank factorization affects LLM trustworthiness across privacy, adversarial robustness, ethics, and fairness, complemented by an explainability-driven analysis of the internal mechanisms behind these trust-related changes. We evaluate multiple LLMs of different sizes and architectures compressed with various low-rank factorization algorithms, revealing key insights: (1) low-rank factorization preserves training data privacy but weakens the protection of personally identifiable information during conversations; (2) adversarial robustness is generally enhanced under compression; (3) ethics degrades in zero-shot prompting but partially recovers in few-shot prompting; (4) fairness declines under compression. Beyond compression, we investigate how model scale and fine-tuning affect trustworthiness. Additionally, to move beyond black-box analysis, we employ a gradient-based attribution to identify which layers of LLMs contribute most to adversarial robustness.
Static Models, Dynamic World: A Unified Perspective on Temporal Perception in Large Language Models
PDF ↗Large language models are trained on static corpora but deployed in a dynamic world, leading to systematic temporal failures—from mis-anchored expressions and inconsistent timelines to hallucinated future events, stale world knowledge, and related issues. Existing surveys on temporal knowledge graphs, retrieval-augmented generation, hallucination, and knowledge editing cover only isolated fragments of this space: they are typically task-centric and do not offer a holistic theoretical account of how frozen LLMs represent and reason about time. This survey provides a unified perspective on temporal reasoning in LLMs. We formalize temporal queries in an information-theoretic framework based on the parametric reachability of temporal premises and answers, which induces four temporal information regimes corresponding to internal reasoning, answer recency, premise anchoring, and genuine world indeterminacy. Under this lens, we delineate the landscape of temporal failure modes, consolidate methodologies for diagnosing temporal deficiencies, and synthesize mitigation approaches into a coherent design space. Together, these contributions provide a systematic roadmap toward reliable time-aware large language models.
Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage
PDF ↗Large language models (LLMs) demonstrate strong capabilities across a wide range of complex tasks and are increasingly deployed at scale, placing significant demands on inference efficiency. Prior work typically decomposes inference into prefill and decode stages, with the decode stage dominating total latency. To reduce time and memory complexity in the decode stage, a line of work introduces sparse-attention algorithms. In this paper, we show, both empirically and theoretically, that sparse attention can paradoxically increase end-to-end complexity: information loss often induces significantly longer sequences, a phenomenon we term “Less is Less” (Lil). To mitigate the Lil problem, we propose an early-stopping algorithm that detects the threshold where information loss exceeds information gain during sparse decoding. Our early-stopping algorithm reduces token consumption by up to 90% with a marginal accuracy degradation of less than 2% across reasoning-intensive benchmarks.
Recent advancements in Spatial Intelligence (SI) have predominantly relied on Vision-Language Models (VLMs), yet a critical question remains: does spatial understanding originate from visual encoders or the fundamental reasoning backbone? Inspired by this question, we introduce **SiT-Bench**, a novel benchmark designed to evaluate the SI performance of Large Language Models (LLMs) without pixel-level input, comprises over 3,800 expert-annotated items across five primary categories and 17 subtasks, ranging from egocentric navigation and perspective transformation to fine-grained robotic manipulation. By converting single/multi-view scenes into high-fidelity, coordinate-aware textual descriptions, we challenge LLMs to perform symbolic textual reasoning rather than visual pattern matching. Evaluation results of state-of-the-art (SOTA) LLMs reveals that while models achieve proficiency in localized semantic tasks, a significant "spatial gap" remains in global consistency. Notably, we find that explicit spatial reasoning significantly boosts performance, suggesting that LLMs possess latent world-modeling potential. Our proposed dataset SiT-Bench serves as a foundational resource to foster the development of spatially-grounded LLM backbones for future VLMs and embodied agents.
Multi-hop Knowledge Graph Question Answering (KGQA) requires coherent reasoning across relational paths, yet existing methods often treat each reasoning step independently and fail to effectively leverage experience from prior explorations, leading to fragmented reasoning and redundant exploration. To address these challenges, we propose Trajectoryaware Reasoning with Adaptive Context and Exploration priors (TRACE), an experiential framework that unifies LLM-driven contextual reasoning with exploration prior integration to enhance the coherence and robustness of multihop KGQA. Specifically, TRACE dynamically translates evolving reasoning paths into natural language narratives to maintain semantic continuity, while abstracting prior exploration trajectories into reusable experiential priors that capture recurring exploration patterns. A dualfeedback re-ranking mechanism further integrates contextual narratives with exploration priors to guide relation selection during reasoning. Extensive experiments on multiple KGQA benchmarks demonstrate that TRACE consistently outperforms state-of-the-art baselines.
A Multilingual Dataset and Empirical Validation for the Mutual Reinforcement Effect in Information Extraction
PDF ↗The Mutual Reinforcement Effect (MRE) describes a phenomenon in information extraction where word-level and sentence-level tasks can mutually improve each other when jointly modeled. While prior work has reported MRE in Japanese, its generality across languages and task settings has not been empirically validated, largely due to the lack of multilingual MRE datasets. To address this limitation, we introduce the Multilingual MRE Mix dataset (MMM), which consists of 21 sub-datasets covering English, Japanese, and Chinese. We propose an LLM-assisted dataset translation and alignment framework that significantly reduces manual annotation effort while preserving the structural requirements of MRE tasks. Building on MMM, we adopt a unified input-output framework to train an open-domain information extraction model and conduct extensive empirical studies, including full fine-tuning ablations and the construction of knowledgeable verbalizers based on MRE-mix data. Experimental results show that 76 percent of the MMM sub-datasets consistently exhibit the Mutual Reinforcement Effect across languages. These findings provide systematic empirical validation of MRE in multilingual settings and demonstrate its practical value for information extraction.
Role-playing models (RPMs) are widely used in real-world applications but underperform when deployed in the wild. This degradation can be attributed to distribution shifts, including user, character, and dialogue compositional shifts. Existing methods like LLM-as-a-judge fall short in providing a fine-grained diagnosis of how these shifts affect RPM generalization, and thus there lack formal frameworks to characterize RPM generalization behaviors. To bridge these gaps, we introduce an information-theoretic metric, named reasoning-based effective mutual information difference (R-EMID), to measure RPM performance degradation in an interpretable way. We also derive an upper bound on R-EMID to predict the worst-case generalization performance of RPMs and theoretically reveal how various shifts contribute to the RPM performance degradation. Moreover, we propose a co-evolving reinforcement learning framework to adaptively model the connection among user, character, and dialogue context and thus enhance the estimation of dialogue response generation probability, which is critical for calculating R-EMID. Finally, we evaluate the generalization performance of various RPMs using R-EMID, finding that user shift poses the highest risk among all shifts and reinforcement learning is the most effective approach for enhancing RPM generalization. Code and data are available at https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/RPM-Generalization.
Large language models (LLMs) increasingly rely on external tools to complete complex tasks, yet their ability to recognize and correct their own tool-use mistakes remains underexplored. Existing benchmarks primarily evaluate planning and execution success, overlooking the self-reflective dimension of tool use. To address this gap, we present ReflecTool-Bench, the first benchmark designed to assess LLMs’ self-reflective reasoning in tool-augmented multi-turn dialogues. ReflecTool-Bench covers 10 domains with 88 distinct APIs and 968 annotated dialogues, systematically injecting diverse error types arising from both user and assistant behavior. The benchmark defines two complementary evaluation setups: the Critique task, where models diagnose errors in third-party dialogues, and the Self-Reflection Task, where models must detect and repair their own prior tool-use mistakes. We introduce fine-grained metrics for error detection, error classification, correction accuracy, and explanation quality, enabling a holistic assessment of reflective reasoning. Evaluations across 12 state-of-the-art models, including both API-based closed source models and open source models, reveal that while models can reliably identify user-originated errors, they struggle with assistant-originated ones, and performance drops sharply when moving from critique to self-reflection.
Rethinking Research on Stereotypes: An Analysis through Social Psychological and Computational Perspectives
PDF ↗Stereotypes are social constructs shaping human perception and behavior that can produce harmful outcomes under specific conditions. Recent work shows that large language models (LLMs) may inherit and amplify such social harms. However, most existing research often focuses only upon stereotypical biases and overlooks stereotypes and the rich social psychological literature on them, resulting in resource wastage and slowed progress in stereotype research.We argue that meaningful progress in mitigating stereotypes in LLMs requires tighter integration between social psychology and computational research. To address this gap, we review core social psychological theories and frameworks and analyze their computational operationalization, highlighting substantial open opportunities.We also analyze computational progress across media narratives, body imaging, and multilingual, multicultural, and multimodal contexts, identifying key gaps and limitations in each domain.We also present a unified analysis of challenges in stereotype research.We further discuss implications for responsible AI, highlighting stereotypes as a major source of downstream harms, and briefly examine the limitations of current mitigation approaches along with potential improvements via explainability and interpretability. We frame stereotypes in AI as socio-technical phenomena and urge further research in responsible AI, informed by the perspectives and future directions presented in this paper.
BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards
PDF ↗Critic-free reinforcement learning with verifiable rewards (RLVR), exemplified by Group Relative Policy Optimization (GRPO), avoids training a value function (critic) and reduces memory and compute overhead relative to critic-based PPO pipelines for aligning large language models. However, GRPO-style advantage estimation depends on prompt-local (within-prompt-group) reward statistics and can be unstable. In particular, when all rollouts in a prompt group receive identical rewards, the within-group reward variance becomes zero, and group normalization yields *zero* advantages for that group, impeding learning in cold-start regimes with binary verifiers. We introduce **BV-Blend**, a critic-free framework that stabilizes advantage estimation by combining prompt-local on-policy statistics with semantic-cluster-conditioned historical moments. BV-Blend maintains EMA-tracked reward moments for each cluster, derives a confidence weight from a standard error of the mean (SEM) proxy, and uses this weight to blend historical and prompt-local baseline and variance statistics into a standardized advantage for PPO-style clipped updates. Experiments on verifiable reasoning benchmarks show that BV-Blend improves training stability and performance, and remains robust in regimes where group-normalized methods may stall.