AI 速览：提出一种以学习者参与为核心的多智能体组织框架，并在 Agentic University 中实现，用可修订的“能力项目”把个人目标、学习活动与分层评估串联起来。
Large language models support personalized tutoring, content generation, and adaptive learning paths. However, capability development requires these resources and interactions to form a coherent learning process that evolves with the learner. A central challenge is to align goals, activities, and assessment while using learning evidence to decide what should follow. We introduce Human-Participatory Multi-Agent Organization and instantiate it in Agentic University (AU). In AU, learners participate as organization members and work with specialized agents to maintain a revisable Capability Program. The program connects personal goals, learning activities across modalities, and layered assessment, while learner feedback and performance guide subsequent support. Our evaluation shows advantages in program organization, assessment design, and revision quality. Users achieved larger mean assessment gains (12.19 versus 5.61 points on a 50-point scale) and reported stronger cognitive support. This work contributes an organizational approach to adaptive learning that makes learner participation a continuing basis for multi-agent coordination. The Agentic University project is available at \href{https://github.com/AgenticOrgLab/AgenticUniversity.git}.
Chain-of-Thought (CoT) reasoning has significantly enhanced the reasoning capabilities of large language models (LLMs). However, existing reinforcement learning methods primarily optimize outcome-level rewards, providing limited supervision over the reasoning process. As a result, models may produce shortcut-prone or unstable reasoning trajectories even when the final answer is correct. We propose GeoFaith, a spatio-temporal latent dynamics framework for learning faithful reasoning processes without process reward models. GeoFaith combines final-state latent uncertainty, which measures whether a completed trajectory lies in a well-supported manifold region, with predictive-entropy dynamics, which penalize abnormal uncertainty patterns such as premature collapse, oscillation, and prolonged uncertainty. These signals are incorporated as process regularization into GRPO, jointly optimizing answer correctness and reasoning quality without step-level process annotations. Experiments show that GeoFaith consistently improves reasoning accuracy, produces more efficient reasoning trajectories, and outperforms both outcome-only reinforcement learning and process-supervision baselines.
Yunqi Liu, Peiyu Wang, Bingli Wang, Leo Luo, Xuanwu Song, Xudong Ma, Hongxin Li, Yaxin Du 等 24 位作者
Command-line interface (CLI) agents increasingly execute longer-horizon tasks, yet existing benchmarks either assume that user instructions remain fixed throughout execution or evaluate task outcomes without capturing runtime collaboration quality. In this paper, we present REALCLI, a paired benchmark that unifies the evaluation of task execution and human-agent collaboration quality on realistic CLI tasks. To comprehensively measure collaboration, REALCLI builds a five-dimensional framework, operationalized into 22 sub-dimensions and 51 interaction scenario templates. REALCLI provides two modes for paired comparison: REALCLI-EXEC supplies complete specifications upfront, whereas REALCLI-COLLAB introduces a scripted user that delivers progressive information, corrections, constraint revisions, and authorization decisions. Evaluations reveal profound capability gaps in frontier agents: the best configuration attains an average score of only 49.2%, and when evaluated by the joint collaboration score ($J_{\mathrm{collab}}$), the peak performance reaches merely 37.6%. These results show that task effectiveness diverges sharply from collaboration quality, with models consistently struggling in steerability, trust calibration, and safety boundary management. REALCLI provides a rigorous, reproducible foundation for measuring and advancing collaborative CLI agents.
Tianci Gao Renmin University of China, Haoxuan Li Tsinghua University, Jian He Li, Tianxiang Zhao, Shi Runze, Weiran Wang, Zezhao Wu, Lu Mi Tsinghua University
Scientific AI agents can autonomously carry out complex research workflows, yet these unfolded workflows often remains difficult for humans to inspect and review, limiting interpretable, controllable and effective human–AI collaboration. To address this challenge, we present a monitoring and visualization framework that records fine-grained execution events and organizes them into a directed graph that make agent workflows explicit as they proceed. The system records intermediate steps (e.g. tool calls and code executions), and renders them as real-time updated visual traces that expose workflow structure. This allows users to examine how results are produced, identify where failures emerge, and better understand agent behavior across different stages of the research process.We conduct an evaluation on complex research tasks with domain experts of interdisciplinary background in AI, neuroscience and biology. Experts report that structured traces visualization improves understanding of agent workflows, perceived interpretability, and usability for analysis and further interaction.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) · 2026 · 正式发表
Xiaoxiao Ma, Kuofeng Gao Tsinghua University, Zeyi Lu, Wenxi Jiang, Hao Fang Beijing University of Post and Telecommunications and Tsinghua University, Hao Wu Tsinghua University, Tsinghua University, Bin Chen Harbin Institute of Technology, Shenzhen, Shu-Tao Xia Shenzhen International Graduate School, Tsinghua University
Key-Value (KV) caching is widely used in large language models (LLMs) to enable long-context inference efficiently, yet its security implications remain underexplored. We present the first systematic study of how KV cache compression interacts with jailbreak attacks, evaluating four model families under diverse jailbreak attacks. We identify a double-edged effect: (i) on one hand, compression can induce Accidental Robustness, where optimization-based and encoding-based attacks fail due to Malicious Semantic Eviction, where attacks’ own attention redirection reduces the malicious query’s cache importance, and Gradient Mismatch where discrete compression operations break jailbreak optimization. (ii) On the other hand, Vulnerability Paradox arises under merging-based compression for human-designed Attacks, where aggressive merging in shallow layers triggers functional head collapse, amplifying attack success rates. To address this, we propose Safe-CAM, a history-aware, per-head feedback merging strategy that prevents safety degradation while maintaining efficiency. Experiments show Safe-CAM fully restores safety (0% ASR) and improves benign task performance with minimal overhead. Our study highlights that KV cache compression is not only an efficiency mechanism but also a safety-critical design factor in LLM deployment.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Tao Feng Zhejiang University, Xinke Jiang, Xinyan Hu, Yonggang Zhang, Zhen Tao Harbin Institute of Technology, Shenzhen, Wentao Zhang, Boyang Liu, Wenhao Jiang Harbin Institute of Technology 等 9 位作者
Agentic search augments large language models (LLMs) with external knowledge through reinforcement learning. However, existing approaches suffer from blind reliance on noisy retrieval and hallucination when both parametric and external knowledge fail—reflecting a lack of calibration regarding the model’s knowledge boundary. We propose Knowledge boundary Policy Optimization (KbPO), a reinforcement learning framework that explicitly aligns retrieval decisions with quantified knowledge states. KbPO introduces: (1) a semantic stability metric to delineate reliable parametric knowledge; (2) a four-quadrant taxonomy synthesising internal certainty with retrieval quality; and (3) a quadrant-based reward mechanism incentivising calibrated behaviour. We further adopt an iterative query evolution pipeline to construct boundary-probing training samples. Experiments on ten benchmarks demonstrate that KbPO outperforms strong baselines while exhibiting reduced hallucination rates.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Kairui Hu Nanyang Technological University, Penghao Wu Nanyang Technological University, Fanyi Pu, Wang Xiao, Xiang Yue Carnegie Mellon University, Bo Li, Yuanhan Zhang, Ziwei Liu Nanyang Technological University
Humans acquire knowledge through three cognitive stages: perceiving information, comprehending knowledge, and adapting knowledge to solve novel problems. Videos serve as an effective medium for knowledge acquisition, facilitating a natural progression through these learning stages. However, existing video benchmarks fail to evaluate the knowledge acquisition capabilities of Large Multimodal Models (LMMs). To address this gap, we introduce Video-MMMU, a multi-modal, multi-discipline, multi-track benchmark that evaluates LMMs’ ability to acquire knowledge from college-level, educational videos. Video-MMMU features a collection of 300 videos and 900 human-annotated questions across six disciplines, evaluating knowledge acquisition through stage-aligned question-answer pairs: Perception, Comprehension, and Adaptation. Beyond measuring final accuracy, Video-MMMU proposes the performance gain metric that quantifies an LMM’s learning gain from video, shifting the focus of evaluation from absolute performance to learning efficiency. Our evaluation reveals a substantial gap between human learners and current LMMs, highlighting the need to improve models’ ability to learn and adapt knowledge from video content.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Jiayi Zhang, Shu Yang, Junchao Wu University of Macau, Derek F. Wong University of Macau, Di Wang KAUST
Fine-tuning Large Language Models on a political topic will significantly manipulate their political stance on various issues and unintentionally affect their stance on broad topics. While previous studies have proposed this issue, there is still a lack of understanding regarding the internal representations of these stances and the mechanisms that lead to unintended cross-topic generalization. In this paper, we systematically explore the internal mechanisms underlying this phenomenon from a neuron-level perspective and how to mitigate the cross-topic generalization of political fine-tuning. Firstly, we propose Political Neuron Localization through Activation Contrasting (PNLAC) to identify two distinct types of political neurons: general political neurons, which govern stance across multiple political topics, and topic-specific neurons that affect the model’s political stance on individual topics. We find that these political neuron types exist in the middle and later layers across four models and datasets through activation patching experiments. Leveraging these insights, we introduce InhibitFT, an inhibition-based fine-tuning method that effectively mitigates the cross-topic stance generalization. Experimental results demonstrate the robustness of the identified neuron types across various models and datasets and show that InhibitFT significantly reduces the cross-topic stance generalization by 20% on average while preserving topic-specific performance. Moreover, we demonstrate that selectively inhibiting only 5% of neurons is sufficient to effectively mitigate the cross-topic stance generalization.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Jia-Nan Li, Jian Guan, Songhao Wu, Wei Wu Ant Research, Rui Yan Wuhan University
Large language models (LLMs) have traditionally been aligned through one-size-fits-all approaches that assume uniform human preferences, fundamentally overlooking the diversity in user values and needs. This paper introduces a comprehensive framework for scalable personalized alignment of LLMs. We establish a systematic preference space characterizing psychological and behavioral dimensions, alongside diverse persona representations for robust preference inference in real-world scenarios. Building upon this foundation, we introduce AlignX, a large-scale dataset of over 1.3 million personalized preference examples, and develop two complementary alignment approaches: in-context alignment directly conditioning on persona representations and preference-bridged alignment modeling intermediate preference distributions. Extensive experiments demonstrate substantial improvements over existing methods, with an average 17.06% accuracy gain across four benchmarks while exhibiting a strong adaptation capability to novel preferences, robustness to limited user data, and precise preference controllability. These results validate our approach toward user-adaptive AI systems.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhaoheng Huang Alibaba Group, Dacheng Wen, Yutao Zhu, Xiaoying Lian Microsoft, Yushi Liang Microsoft, Kai Hao Microsoft, Nan Li Microsoft, Liangjie Zhang 等 12 位作者
Large language models (LLMs) in retrieval-augmented generation systems can still produce hallucinations, generating content that is unsupported or contradicted by the source texts and undermines reliability. Recent work addressed this problem by training span-level hallucination detectors using reinforcement learning (RL) and chain-of-thought (CoT) reasoning. In this work, we show through error analysis that incorrect predictions by existing reasoning-based detectors are strongly associated with CoT processes that lack explicit grounding in source evidence, particularly when verification steps do not quote or verify claims against the retrieved documents. This behaviour contrasts with human verification practices in benchmarks such as RAGTruth, where evidence quotation is a prerequisite for determining hallucinated spans. Motivated by this observation, we propose an evidence-grounded RL framework, namely RLSeek, to explicitly enforce active evidence seeking during CoT reasoning by requiring quotation of relevant source segments at each verification step. Experiments on the RAGTruth and NewsSum dataset demonstrate consistent improvements in hallucination span detection performance, with limited additional reasoning overhead and improved robustness in out-of-domain settings.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Preni Golazizian, Elnaz Rahmati, Jackson Trager, Zhivar Sourati, Nona Ghazizadeh, Georgios Chochlakis, Jose J. Alcocer Harvard University, Kerby Bennett University of California, Los Angeles 等 20 位作者
Traffic stops are among the most frequent police–civilian interactions, and body-worn cameras (BWCs) provide a unique record of how these encounters unfold. Respect is a central dimension of these interactions, shaping public trust and perceived legitimacy, yet its interpretation is inherently subjective and shaped by lived experience, rendering community-specific perspectives a critical consideration. Leveraging unprecedented access to Los Angeles Police Department BWC footage, we introduce the first large-scale traffic-stop dataset annotated with respect ratings and free-text rationales from multiple perspectives. By sampling annotators from police-affiliated, justice-system-impacted, and non-affiliated Los Angeles residents, we enable the systematic study of perceptual differences across diverse communities. To this end, (i) we develop a domain-specific evaluation rubric grounded in procedural justice theory, LAPD training materials, and extensive fieldwork; (ii) we introduce a criterion-driven preference data construction framework for perspective-consistent alignment, and (ii) we propose a perspective-aware modeling framework that predicts personalized respect ratings and generates annotator-specific rationales for both officers and civilian drivers from traffic-stop transcripts. Across all three annotator groups, our approach improves both rating prediction performance and rationale alignment. Our perspective-aware framework enables law enforcement to better understand diverse community expectations, providing a vital tool for building public trust and procedural legitimacy.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Songhao Wu, Ang Lv, Ruobing Xie Tencent, Samm Sun, Di Wang, Rui Yan Wuhan University, Yankai Lin Renmin University of China
Mixture-of-Experts (MoE) models rely on an external router to assign tokens to experts. This design inherently separates the routing decision from each expert’s internal capabilities, leading to suboptimal performance. In this work, we address this limitation with Union-of-Experts (UoE), an MoE variant that performs "expert-autonomous routing”. The core mechanism of UoE is to pre-designate a minute fraction of neurons within each expert as "routing neurons”. Experts autonomously select relevant tokens by comparing the activation intensity of these neurons, aligning routing decisions with each expert’s functional profile. To prevent the waste of activations from unselected experts’ routing neurons, we aggregate all routing neuron outputs and sum them into the final layer output. This aggregation acts as a novel virtual shared expert whose parameters are distributed across the individual experts, and improves overall parameter efficiency. We pre-train UoE models with up to 3B parameters, demonstrating that they outperform traditional MoEs with matched efficiency. Furthermore, our analysis of the routing neurons provides valuable insights into expert-autonomous selection and the broader routing mechanisms of MoE models.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Junchao Wu University of Macau, Yefeng Liu, Chenyu Zhu, Hao Zhang, Zeyu Wu, Tianqi Shi, Yichao Du, Longyue Wang Alibaba Group 等 11 位作者
The effective detection and governance of Large Language Model (LLM) generated content has become increasingly critical due to the growing risk of misuse. Despite the impressive performance of existing detectors, their reliability and potential in multilingual, real-world scenarios remain largely underexplored.In this study, we introduce DetectRL-X, a comprehensive multilingual benchmark designed to evaluate advanced detectors across 8 dimensions. The benchmark encompasses 8 languages commonly used in commercial contexts and collects human-written texts from 6 domains highly susceptible to LLM misuse. To better aligned with real-world applications, We create LLM-generated texts using 4 popular commercial LLMs, and include typical AI-assisted writing operations such as polishing, expanding, and condensing to capture authentic usage patterns. Furthermore, we develop a multilingual framework for paraphrasing and perturbation attacks to simulate diverse human modifications and writing noise, enabling stress testing of detectors across languages.Experimental results on DetectRL-X reveal the strengths and limitations of current state-of-the-art detectors when applied to diverse linguistic resources. We further analyze how domains, generators, attack strategies, text length, and refinement operations influence performance in different languages, underscoring DetectRL-X as an effective benchmark for strengthening multilingual and language-specific detectors.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Hao Wu, Hongru Sun University of Wollongong, Wanqing Li University of Wollongong, Xinguo Yu, Hao Ming, Xiao Luo University of Wisconsin - Madison, Wenbin Zhang Florida International University, Jiahong Zhao University of Southampton 等 10 位作者
Mathematical reasoning is one of the core capabilities for Large Language Models (LLMs). Yet, existing approaches often rely on static heuristics or pre-determined reasoning strategies, limiting their ability to adapt to different intermediate states. To address this limitation, we propose FLAIR (Fuzzy-Logic-AssIsted Reasoner), an adaptive framework that integrates fuzzy theory into LLM-based mathematical reasoning. Specifically, FLAIR characterizes intermediate problem-solving states using fuzzy memberships and employs a parameterized fuzzy rule system to conditionally activate subsequent actions. These rule parameters are further adjusted via Reinforcement Learning using solution-level feedback as the reward signal, enabling adaptive and iterative refinement without reliance on a fixed strategy. To the best of our knowledge, this work is the first to integrate fuzzy theory into LLM-based mathematical reasoning. Extensive experiments across multiple benchmarks demonstrate that FLAIR consistently outperforms recent state-of-the-art baselines, while offering effective and interpretable diagnostics of intermediate problem-solving states.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yueqi Xie, Tao Qi Beijing University of Posts and Telecommunications, Jingwei Yi Beijing Academy of Artificial Intelligence, Xiyuan Yang, Ryan Whalen University of Hong Kong, Junming Huang Princeton University, Qian Ding Tsinghua University and Microsoft, Yu Xie 等 10 位作者
With the growing prevalence of generative AI, an increasing amount of content is no longer exclusively generated by humans but by generative AI models with human guidance. This shift presents notable challenges for the delineation of originality due to the varying degrees of human contribution in AI-assisted works. This study raises the research question of measuring human contribution in AI-assisted content generation and introduces a framework to address this question that is grounded in information theory. By calculating mutual information between human input and AI-assisted output relative to self-information of AI-assisted output, we quantify the proportional information contribution of humans in content generation. Our experimental results demonstrate that the proposed measure effectively discriminates between varying degrees of human contribution across multiple creative domains. To further enhance real-world applicability, we extend the framework to estimate the minimal necessary human contribution for any text without requiring human input and validate its effectiveness. We hope that this work lays a foundation for measuring human contributions in AI-assisted content generation in the era of generative AI.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Shanwen Tan Sichuan University, Ziyang Dong, Wei Ju Sichuan University and Peking University, Yiwei Fu, Hao Wu Tencent, Kun Wang, Yifan Wang University of International Business and Economics, Ziyue Qiao Great Bay University
This paper investigates the problem of safe decoding for Large Language Models (LLMs) during inference, particularly under jailbreak attacks. Previous approaches typically either detect malicious content or regulate the decoding alignment of LLMs to mitigate such attacks. Although effective in defending against attacks, these methods often over-reject benign content, limiting their generalizability in real-world scenarios where harmful and benign information coexist. Towards this end, we propose an innovative framework named Sequence-level risk Accumulation for calibrating test-time alignment (SEAT). Specifically, SEAT introduces a reward-guided branch decoding paradigm to incorporate safety awareness during generation. To balance the detection of harmful content with the accurate response to benign information, SEAT employs a sequence-level risk monitor that smooths risk signals over the entire sequence, preventing over-confident refusals for certain tokens. Furthermore, we conduct extensive experiments on four attack benchmarks and two neutral datasets, comparing SEAT with eight state-of-the-art baselines. Consequently, the results demonstrate that SEAT achieves superior performance both in defending against jailbreak attacks and in generating high-quality responses on neutral datasets. Our code is available at https://github.com/ShanwenTan/SEAT.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Evaluating expressive speech remains challenging, as existing methods mainly assess emotional intensity and overlook whether a speech sample is expressively appropriate for its contextual setting. This limitation hinders reliable evaluation of speech systems used in narrative-driven and interactive applications, such as audiobooks and conversational agents. We introduce CEAEval, a Context-rich framework for Evaluating Expressive Appropriateness in speech, which assesses whether a speech sample expressively aligns with the underlying communicative intent implied by its discourse-level narrative context. To support this task, we construct CEAEval-D, the first context-rich speech dataset with real human performances in Mandarin conversational speech, providing narrative descriptions together with fifteen dimensions of human annotations covering expressive attributes and expressive appropriateness. We further develop CEAEval-M, a model that integrates knowledge distillation, planner-based multi-model collaboration, adaptive audio attention bias, and reinforcement learning to perform context-rich expressive appropriateness evaluation. Experiments on a human-annotated test set demonstrate that CEAEval-M substantially outperforms existing speech evaluation and analysis systems.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Sunguk Shin, Fangzhao Wu Microsoft, Byung-Jun Lee Korea University and Gauss Labs Inc., Meeyoung Cha Max Planck Institute for Security and Privacy and Korea Advanced Institute of Science & Technology, Sungwon Park
Open-weight large language models (LLMs) enable broad customization, but also increase exposure to post-release misuse, including malicious fine-tuning (MFT). To mitigate this risk, many prior defenses aim to improve the robustness of open-weight models to MFT by constraining adversarial fine-tuning dynamics in parameter space or mitigating harmful information encoded in internal representations. Nevertheless, since malicious fine-tuning can still erode safety, developing robust safeguards for open-weight models that fundamentally mitigate this risk remains an open research problem. In this paper, we characterize a safety region for open-weight LLMs and propose Safety Guidance Trigger (SGT), which guides fine-tuning toward the safety manifold to preserve alignment. SGT has two stages: (1) optimizing a safety trigger that steers the base model toward safe responses and (2) training the open-weight model to align its internal features with trigger-induced safety representations. We demonstrate that SGT substantially improves robustness against malicious fine-tuning, requiring adversaries to increase their data budget significantly to compromise safety. Our analysis shows that SGT anchors model representations to a safety region, which remains stable under malicious fine-tuning.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Haiqing Li, Wenliang Zhong University of Texas at Arlington, Yinhao Wu, Hehuan Ma University of Texas at Arlington, Yuzhi Guo University of Texas at Arlington, Thao M. Dang University of Texas at Arlington, Junzhou Huang University of Texas, Arlington
Guideline-following is increasingly important in compliance, customer support, and other regulated workflows, where correctness is defined by explicit rule systems rather than heuristics. Learning to follow guidelines is challenging because guidelines are interdependent: rules can trigger, suppress, or conflict with one another, while locally plausible responses may violate global constraints. Most existing methods treat guidelines as static text and rely on implicit reasoning or deeper decoding, making rule interactions and satisfaction status hard to observe and control. A more feasible approach is to model guideline execution with an explicit state that tracks evolving rule evidence across steps. However, conventional world models are a poor fit: they typically assume privileged feedback or well-defined transition dynamics, assumptions that do not hold when reasoning occurs purely in language space under ambiguous, text-defined constraints. As a solution, we propose RGCWM, a Rule-Grounded Causal World Model that builds an explicit state space from the guideline text itself. RGCWM represents rule applicability and satisfaction as a continuously updated evidence state, externalizes inter-rule dependencies as a causal structure, and plans at inference time by counterfactually evaluating candidate responses under model-estimated state transitions. Experiments show that this shift from implicit text reasoning to state-based reasoning enables stable, controllable execution of complex interacting rules across diverse domains.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li 等 13 位作者
Evaluating large language model (LLM) based chat assistants is challenging due to their broad capabilities and the inadequacy of existing benchmarks in measuring human preferences. To address this, we explore using strong LLMs as judges to evaluate these models on more open-ended questions. We examine the usage and limitations of LLM-as-a-judge, including position, verbosity, and self-enhancement biases, as well as limited reasoning ability, and propose solutions to mitigate some of them. We then verify the agreement between LLM judges and human preferences by introducing two benchmarks: MT-bench, a multi-turn question set; and Chatbot Arena, a crowdsourced battle platform. Our results reveal that strong LLM judges like GPT-4 can match both controlled and crowdsourced human preferences well, achieving over 80% agreement, the same level of agreement between humans. Hence, LLM-as-a-judge is a scalable and explainable way to approximate human preferences, which are otherwise very expensive to obtain. Additionally, we show our benchmark and traditional benchmarks complement each other by evaluating several variants of LLaMA and Vicuna. The MT-bench questions, 3K expert votes, and 30K conversations with human preferences are publicly available at https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge.