语言大模型 / LLM - arXivDaily 专题

2310.18152 2024-03-12 cs.CL cs.LG 90%

Disentangled Representation Learning with Large Language Models for Text-Attributed Graphs

Yijian Qin, Xin Wang, Ziwei Zhang, Wenwu Zhu

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2403.02509 2024-03-06 cs.CL cs.AI 90%

SPUQ: Perturbation-Based Uncertainty Quantification for Large Language Models

Xiang Gao, Jiaxin Zhang, Lalla Mouatadid, Kamalika Das

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

Comments Accepted to appear at EACL 2024

2402.14404 2024-02-27 cs.CL cs.AI 90%

On the Tip of the Tongue: Analyzing Conceptual Representation in Large Language Models with Reverse-Dictionary Probe

Ningyu Xu, Qi Zhang, Menghan Zhang, Peng Qian, Xuanjing Huang

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；prompting(abstract)；分类 cs.CL、cs.AI

Comments 21 pages, 13 figures

2402.05457 2024-02-09 cs.CL cs.AI cs.MM cs.SD eess.AS 90%

It's Never Too Late: Fusing Acoustic Information into Large Language Models for Automatic Speech Recognition

Chen Chen, Ruizhe Li, Yuchen Hu, Sabato Marco Siniscalchi, Pin-Yu Chen, Ensiong Chng, Chao-Han Huck Yang

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

Comments Accepted to ICLR 2024, 17 pages. This work will be open sourced under MIT license

2311.05085 2024-02-02 cs.CL cs.AI 90%

Characterizing Large Language Models as Rationalizers of Knowledge-intensive Tasks

Aditi Mishra, Sajjadur Rahman, Hannah Kim, Kushan Mitra, Estevam Hruschka

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

2309.06453 2023-12-20 cs.CL cs.LG 90%

Narrowing the Gap between Supervised and Unsupervised Sentence Representation Learning with Large Language Model

Mingxin Li, Richong Zhang, Zhijie Nie, Yongyi Mao

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

Comments Accepted at AAAI24

2311.06330 2023-12-19 cs.AI cs.CE cs.CL cs.MA econ.GN q-fin.EC 90%

Smart Agent-Based Modeling: On the Use of Large Language Models in Computer Simulations

Zengqing Wu, Run Peng, Xu Han, Shuyuan Zheng, Yixin Zhang, Chuan Xiao

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

Comments Source codes are available at https://github.com/Roihn/SABM

2311.01866 2023-11-06 cs.CL cs.AI 90%

Towards Concept-Aware Large Language Models

Chen Shani, Jilles Vreeken, Dafna Shahaf

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；pretraining(abstract)；分类 cs.CL、cs.AI

Comments EMNLP 2023 findings long paper

2310.16523 2023-10-26 cs.CL cs.AI 90%

Improving Diversity of Demographic Representation in Large Language Models via Collective-Critiques and Self-Voting

Preethi Lahoti, Nicholas Blumm, Xiao Ma, Raghavendra Kotikalapudi, Sahitya Potluri, Qijun Tan, Hansa Srinivasan, Ben Packer, Ahmad Beirami, Alex Beutel, Jilin Chen

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；prompting(abstract)；分类 cs.CL、cs.AI

Comments To appear at EMNLP 2023 main conference

2310.08780 2023-10-16 cs.CL cs.AI 90%

"Im not Racist but...": Discovering Bias in the Internal Knowledge of Large Language Models

Abel Salinas, Louis Penafiel, Robert McCormack, Fred Morstatter

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

Comments Warning: This paper discusses and contains content that is offensive or upsetting

2306.17089 2023-09-21 cs.LG cs.CL 90%

Concept-Oriented Deep Learning with Large Language Models

Daniel T. Chang

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2309.05217 2023-09-12 cs.AI cs.CL 90%

Quantifying and Attributing the Hallucination of Large Language Models via Association Analysis

Li Du, Yequan Wang, Xingrun Xing, Yiqun Ya, Xiang Li, Xin Jiang, Xuezhi Fang

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；pretraining(abstract)；分类 cs.CL、cs.AI

2309.01245 2023-09-06 cs.CL cs.AI 90%

Representations Matter: Embedding Modes of Large Language Models using Dynamic Mode Decomposition

Mohamed Akrout

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

2306.01545 2023-06-16 cs.CL cs.AI cs.CR 90%

PassGPT: Password Modeling and (Guided) Generation with Large Language Models

Javier Rando, Fernando Perez-Cruz, Briland Hitaj

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

2510.20129 2026-05-12 cs.CR cs.AI 90%

SAID: Safety-Aware Intent Defense via Prefix Probing for Large Language Models

SAID: 通过前缀探测实现大型语言模型的安全意识意图防御

Yulong Chen, Qi Zhang, Jiawen Zhang, Yadong Liu, Mu Li, Jie Wen, Yong Xu

机构 * Shenzhen Key Laboratory of Visual Object Detection and Recognition, Harbin Institute of Technology, Harbin Institute of Technology, Shenzhen, Shenzhen, China（深圳视觉目标检测与识别重点实验室，哈尔滨工业大学，哈尔滨工业大学，深圳，深圳，中国）

专题命中知识编辑与模型理解：large language model(title,abstract)；language model(title,abstract)；LLM(abstract_cn,comments)；分类 cs.AI

AI总结本文提出SAID框架，通过意图层面的安全探测实现无训练的对抗防御，通过核心意图提炼和安全前缀探测减少有害响应，同时保持良性任务的实用性。

Comments 12 pages, 5 figures. V2: Updated title, author list, and extensive experiments; expanded background on LLM security applications

2504.16833 2026-08-17 cs.SE 版本更新 89%

LRASGen: LLM-based RESTful API Specification Generation

LRASGen：基于大语言模型的RESTful API规范生成

Sida Deng, Rubing Huang, Man Zhang, Chenhui Cui, Dave Towey, Rongcun Wang

专题命中知识编辑与模型理解：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结该研究提出LRASGen，即首个结合LLM与API源代码生成RESTful API的OAS的方法，在20个真实API上的实验显示其生成的规范覆盖更多缺失实体，准确性较好。

Comments Accepted for publication in ACM Transactions on Software Engineering and Methodology (TOSEM). DOI: 10.1145/3810241

2608.11552 2026-08-13 cs.CL cs.AI cs.LG 新提交 89%

Beyond Single-Turn Confidence: Trajectory-Adapted Uncertainty Quantification for LLM Agents

超越单轮置信度：面向大语言模型智能体的轨迹适配不确定性量化

Dylan Bouchard, Mohit Singh Chauhan

专题命中知识编辑与模型理解：LLM(title,summary_cn)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结本研究探讨单轮UQ方法能否迁移至LLM智能体的交互式轨迹场景，评估三类UQ方法在多轮工具使用数据集上的表现，发现黑盒自一致性通常最优，提示需在轨迹层面重新验证单轮UQ方法。

2607.27429 2026-07-31 cs.MA 新提交 89%

Auditing Emergent LLM-Agent Collaboration through Cooperation-Obligation Coupling

通过合作-义务耦合对大语言模型智能体（LLM-agent）的涌现协作进行审计

Zuyuan Zhang, Hanqing Yang, Carlee Joe-Wong, Tian Lan

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结针对LLM-agent协作审计的可审计性缺口，本文提出iCORE表示方法，建立合作-义务耦合审计框架，实验验证其可准确检测缺陷并显著提升轨迹与终端性能。

2607.12624 2026-07-15 cs.CR 新提交 89%

PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis

PVDetector：通过策略违规概念分析检测针对特定目的大语言模型代理的提示注入攻击

Junhui Wang, Hangtao Zhang, Zhirun Zheng, Li Zeng, Jiejun Xiao, Xi Luo, Lihua Yin, Saiqin Long

专题命中知识编辑与模型理解：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结研究针对特定目的LLM代理的提示注入攻击，提出PVDetector框架，通过测量与离线派生的PV概念的隐藏状态对齐来检测攻击，实验表明该方法误报率低、开销小，性能优于现有方法。

Comments Accepted to ACM MM 2026. Code: https://github.com/Claresigle/PVDetector

2606.28229 2026-07-02 cs.SE 新提交 89%

Humanizing Automatically Generated Unit Test Suites with LLM-Based Refactoring

基于LLM重构的自动生成单元测试套件人性化

Wendkûuni C. Ouédraogo, Yinghua Li, Xueqi Dang, Paweł Borsukiewicz, Liang Xiao, Lingfeng Bao, Anil Koyuncu, Jacques Klein, David Lo, Tegawendé F. Bissyandé

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结提出TestHumanizer混合方法，利用LLM对SBST生成的测试套件进行受控重构，提升可读性和可维护性，同时保持编译和覆盖率，实验表明重构后编译率接近100%，可读性显著提升。

2606.24609 2026-06-24 eess.SY cs.SY 新提交 89%

CONDUCTOR: An LLM-Orchestrated Digital Twin for Uncertainty-Aware Distribution Grid Operations

CONDUCTOR: 一种用于不确定性感知配电网运行的LLM编排数字孪生

Antonio Alcántara, Aysegül Kahraman, Anosh Arshad Sundhu, Spyros Chatzivasileiadis

专题命中知识编辑与模型理解：LLM(title,title_cn)；large language model(abstract)；language model(abstract)

AI总结提出CONDUCTOR，一种基于大语言模型编排的数字孪生系统，用于配电网运行，支持不确定性感知分析（概率安全评估、鲁棒校正调度等），在真实丹麦博恩霍尔姆60 kV配电网测试中，首次任务正确率达98.5%。

2606.10210 2026-06-10 cs.CY 新提交 89%

AnnotateThis: Analyzing a human-LLM system for annotating social media data with the concept of climate change mitigation pessimism

AnnotateThis：分析用于以气候变化缓解悲观主义概念标注社交媒体数据的人机协作系统

Zexuan Li, Derek Van Berkel, Ariel Hasell, Grant Schoenebeck, John Barry Ryan, Sabina Tomkins

专题命中知识编辑与模型理解：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结提出人机协作系统AnnotateThis，通过交互式特征帮助用户改进LLM对复杂社会概念的标注质量，实验表明人工干预显著提升F1和准确率。

2606.07937 2026-06-09 cs.CR 新提交 89%

Hallucination Cascade: Analyzing Error Propagation in Multi-Agent LLM Systems

幻觉级联：多智能体大语言模型系统中的错误传播分析

Saeid Jamshidi, Arghavan Moradi Dakhel, Kawser Wazed Nafi, Foutse Khomh

专题命中知识编辑与模型理解：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结本文通过追踪多智能体LLM级联中的声明级事实不一致性，分析幻觉动态，发现更深级联降低幻觉分数但牺牲事实准确性，并揭示模型与领域差异。

2606.06048 2026-06-08 cs.CV 版本更新 89%

LLM-Conditioned Synthesis of Pathological Gaits via Structured Gait-Language Representations

基于结构化步态-语言表示的LLM条件病理步态合成

Mritula Chandrasekaran, Sanket Kachole, Jarek Francik, Dimitrios Makris

机构 * University of California, Berkeley（加州大学伯克利分校）； Stanford University（斯坦福大学）； University of Toronto（多伦多大学）； MIT Media Lab（麻省理工学院媒体实验室）

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结提出一种多模态LLM引导框架，通过结构化文本描述合成病理步态3D数据，利用运动标记化、病理感知语言条件、LLM语义增强和语言到步态生成，改善下游分类性能。

Comments Accepted at CVPR MOMA Workshop 2026 and selected for spotlight presentation at the workshop

2605.29864 2026-05-29 cs.RO 89%

LLM-Guided Future Hypotheses for Horizon-Aware Exploration in Multi-Step Robot Manipulation

LLM引导的未来假设用于多步机器人操作中的视野感知探索

Mohammad Khoshnazar, Andrew Melnik, Michael Beetz

机构 * Institute of Artificial Intelligence, University of Bremen（人工智能研究所，不莱梅大学）

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结提出未来经验条件化（FEC）框架，利用LLM生成短期未来视频作为结构化先验，结合行为克隆和强化学习微调，提升多步机器人操作中的探索和策略适应能力。

2605.28571 2026-05-28 cs.HC 89%

Not All Uncertainty Is Equal: How Uncertainty Granularity Shapes Human Verification in LLM-Assisted Decision Making

并非所有不确定性都相同：不确定性粒度如何影响LLM辅助决策中的人类验证

Mauricio Villavicencio, Sitong Pan, Qianwen Wang

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结本研究通过大规模用户实验，探究LLM输出中不同粒度（输出级、关系级、词元级）的不确定性表达对用户验证行为的影响，发现词元级不确定性增加用户对AI的认同，而关系级不确定性减少外部验证。

Comments 54 pages, 36 figures, accepted by ACM FAccT 2026

2605.15455 2026-05-18 cs.HC 89%

Multi-Turn Neural Transparency: Surfacing Neural Activations Improves User Calibration to LLM Behavioral Drift

多轮神经透明：揭示神经激活以改善用户对LLM行为漂移的校准

Sheer Karny, Anthony Baez, Pat Pataranutaporn

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结本文提出多轮神经透明方法，通过实时展示LLM内部激活以帮助用户预测和识别行为变化，实验表明该方法显著提升了用户对模型行为的评估和校准。

2605.01047 2026-05-05 cs.CR cs.AI cs.CL cs.LG 89%

LLM Ghostbusters: Surgical Hallucination Suppression via Adaptive Unlearning

LLM Ghostbusters: 通过自适应反学习实现精准幻觉抑制

Joseph Spracklen, Pedram Aghazadeh, Farinaz Koushanfar, Murtuza Jadliwala

机构 * University of Texas San Antonio（德克萨斯大学圣安东尼奥分校）； University of California San Diego（加州大学圣地亚哥分校）

专题命中知识编辑与模型理解：LLM(title,title_cn)；分类 cs.CL、cs.AI、cs.LG

AI总结本文提出自适应反学习框架，通过混合token级目标和自适应发现循环，有效抑制LLM幻觉，减少81%的包幻觉率，同时保持编码基准性能，验证其对特定分布的针对性效果。

2511.22409 2026-05-04 cs.SE 89%

NOMAD: A Multi-Agent LLM System for UML Class Diagram Generation from Natural Language Requirements

NOMAD：一种多智能体LLM系统，用于从自然语言需求生成UML类图

Polydoros Giannouris, Sophia Ananiadou

专题命中知识编辑与模型理解：LLM(title,title_cn)；large language model(abstract)；language model(abstract)

AI总结 NOMAD通过分解UML生成为多个角色专用子任务，提升可解释性并支持针对性验证。通过北风案例和人工UML练习评估，NOMAD在生成质量上优于基线模型，并揭示了细粒度属性提取的持续挑战。

2604.27714 2026-05-01 cs.CR cs.SE 89%

How Code Representation Shapes False-Positive Dynamics in Cross-Language LLM Vulnerability Detection

代码表示如何影响跨语言LLM漏洞检测中的假阳性动态

Maofei Chen, Laifu Wang, Yue Qin, Yuan Wang, Bo Wu, Dongxin Liu

专题命中知识编辑与模型理解：LLM(title,title_cn)

AI总结本文研究了代码表示格式对跨语言LLM漏洞检测假阳性的影响，通过对比不同训练强度和代码表示格式，发现文本微调会提高假阳性率，而AST表示能有效降低假阳性，且在不同语言间具有鲁棒性。

Comments 14 pages, 5 figures

1. 知识编辑与模型理解 7552 篇

Disentangled Representation Learning with Large Language Models for Text-Attributed Graphs

SPUQ: Perturbation-Based Uncertainty Quantification for Large Language Models

On the Tip of the Tongue: Analyzing Conceptual Representation in Large Language Models with Reverse-Dictionary Probe

It's Never Too Late: Fusing Acoustic Information into Large Language Models for Automatic Speech Recognition

Characterizing Large Language Models as Rationalizers of Knowledge-intensive Tasks

Narrowing the Gap between Supervised and Unsupervised Sentence Representation Learning with Large Language Model

Smart Agent-Based Modeling: On the Use of Large Language Models in Computer Simulations

Towards Concept-Aware Large Language Models

Improving Diversity of Demographic Representation in Large Language Models via Collective-Critiques and Self-Voting

"Im not Racist but...": Discovering Bias in the Internal Knowledge of Large Language Models

Concept-Oriented Deep Learning with Large Language Models

Quantifying and Attributing the Hallucination of Large Language Models via Association Analysis

Representations Matter: Embedding Modes of Large Language Models using Dynamic Mode Decomposition

PassGPT: Password Modeling and (Guided) Generation with Large Language Models

SAID: Safety-Aware Intent Defense via Prefix Probing for Large Language Models

LRASGen: LLM-based RESTful API Specification Generation

Beyond Single-Turn Confidence: Trajectory-Adapted Uncertainty Quantification for LLM Agents

Auditing Emergent LLM-Agent Collaboration through Cooperation-Obligation Coupling

PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis

Humanizing Automatically Generated Unit Test Suites with LLM-Based Refactoring

CONDUCTOR: An LLM-Orchestrated Digital Twin for Uncertainty-Aware Distribution Grid Operations

AnnotateThis: Analyzing a human-LLM system for annotating social media data with the concept of climate change mitigation pessimism

Hallucination Cascade: Analyzing Error Propagation in Multi-Agent LLM Systems

LLM-Conditioned Synthesis of Pathological Gaits via Structured Gait-Language Representations

LLM-Guided Future Hypotheses for Horizon-Aware Exploration in Multi-Step Robot Manipulation

Not All Uncertainty Is Equal: How Uncertainty Granularity Shapes Human Verification in LLM-Assisted Decision Making

Multi-Turn Neural Transparency: Surfacing Neural Activations Improves User Calibration to LLM Behavioral Drift

LLM Ghostbusters: Surgical Hallucination Suppression via Adaptive Unlearning

NOMAD: A Multi-Agent LLM System for UML Class Diagram Generation from Natural Language Requirements

How Code Representation Shapes False-Positive Dynamics in Cross-Language LLM Vulnerability Detection