语言大模型 / LLM - arXivDaily 专题

2407.07071 2024-10-04 cs.CL cs.AI cs.LG 90%

Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Maps

Yung-Sung Chuang, Linlu Qiu, Cheng-Yu Hsieh, Ranjay Krishna, Yoon Kim, James Glass

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、cs.LG

Comments EMNLP 2024 main conference long paper. The source code is available at https://github.com/voidism/Lookback-Lens

2409.13714 2024-09-24 cs.CL cs.AI cs.LG 90%

TracrBench: Generating Interpretability Testbeds with Large Language Models

Hannes Thurnherr, Jérémy Scheurer

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、cs.LG

Comments 6 pages + appendix, 4 figures, ICML Mechanistic Interpretability Workshop

2406.10254 2024-09-19 cs.CL cs.AI cs.LG cs.SD eess.AS 90%

Towards Signal Processing In Large Language Models

Prateek Verma, Mert Pilanci

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、cs.LG

Comments 12 pages, 3 figures

2405.20797 2024-06-18 cs.CV cs.AI cs.CL cs.LG 90%

Ovis: Structural Embedding Alignment for Multimodal Large Language Model

Shiyin Lu, Yang Li, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang, Han-Jia Ye

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、cs.LG

2403.04696 2024-06-10 cs.CL cs.AI cs.LG 90%

Fact-Checking the Output of Large Language Models via Token-Level Uncertainty Quantification

Ekaterina Fadeeva, Aleksandr Rubashevskii, Artem Shelmanov, Sergey Petrakov, Haonan Li, Hamdy Mubarak, Evgenii Tsymbalov, Gleb Kuzmin, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, Maxim Panov

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、cs.LG

Comments Accepted to ACL-2024 (Findings). Ekaterina Fadeeva, Aleksandr Rubashevskii, and Artem Shelmanov contributed equally

2311.15983 2024-06-06 cs.LG cs.AI cs.CL 90%

SPIN: Sparsifying and Integrating Internal Neurons in Large Language Models for Text Classification

Difan Jiao, Yilun Liu, Zhenwei Tang, Daniel Matter, Jürgen Pfeffer, Ashton Anderson

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、cs.LG

Comments 17 pages, 7 figures, 12 tables Code available at https://github.com/difanj0713/SPIN

2307.01379 2024-05-30 cs.CL cs.AI cs.LG 90%

Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models

Jinhao Duan, Hao Cheng, Shiqi Wang, Alex Zavalny, Chenan Wang, Renjing Xu, Bhavya Kailkhura, Kaidi Xu

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、cs.LG

Comments To appear in ACL 2024

2403.14472 2024-05-29 cs.CL cs.AI cs.CV cs.HC cs.LG 90%

Detoxifying Large Language Models via Knowledge Editing

Mengru Wang, Ningyu Zhang, Ziwen Xu, Zekun Xi, Shumin Deng, Yunzhi Yao, Qishen Zhang, Linyi Yang, Jindong Wang, Huajun Chen

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

Comments ACL 2024. Project website: https://zjunlp.github.io/project/SafeEdit Benchmark: https://huggingface.co/datasets/zjunlp/SafeEdit

2402.16123 2024-04-30 cs.CL cs.AI cs.CV cs.HC cs.LG 90%

InstructEdit: Instruction-based Knowledge Editing for Large Language Models

Ningyu Zhang, Bozhong Tian, Siyuan Cheng, Xiaozhuan Liang, Yi Hu, Kouying Xue, Yanjie Gou, Xi Chen, Huajun Chen

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、cs.LG

Comments IJCAI 2024; the project website is at https://www.zjukg.org/project/InstructEdit/

2401.12576 2024-04-25 cs.CL cs.AI cs.LG 90%

LLMCheckup: Conversational Examination of Large Language Models via Interpretability Tools and Self-Explanations

Qianli Wang, Tatiana Anikina, Nils Feldhus, Josef van Genabith, Leonhard Hennig, Sebastian Möller

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、cs.LG

Comments Accepted to NAACL 2024 HCI+NLP workshop; camera-ready version

2404.13044 2024-04-22 cs.CV 90%

Unified Scene Representation and Reconstruction for 3D Large Language Models

Tao Chu, Pan Zhang, Xiaoyi Dong, Yuhang Zang, Qiong Liu, Jiaqi Wang

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

Comments Project Page: https://chtsy.github.io/uni3drr-page/

2309.03883 2024-03-12 cs.CL cs.AI cs.LG 90%

DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

Yung-Sung Chuang, Yujia Xie, Hongyin Luo, Yoon Kim, James Glass, Pengcheng He

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、cs.LG

Comments ICLR 2024 main conference paper. The source code is available at https://github.com/voidism/DoLa

2402.01761 2024-02-06 cs.CL cs.AI cs.LG 90%

Rethinking Interpretability in the Era of Large Language Models

Chandan Singh, Jeevana Priya Inala, Michel Galley, Rich Caruana, Jianfeng Gao

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、cs.LG

Comments 7 pages

2309.01029 2023-11-30 cs.CL cs.AI cs.LG 90%

Explainability for Large Language Models: A Survey

Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Mengnan Du

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、cs.LG

2304.02754 2023-11-13 cs.AI cs.CL cs.LG 90%

Conceptual structure coheres in human cognition but not in large language models

Siddharth Suresh, Kushin Mukherjee, Xizheng Yu, Wei-Chun Huang, Lisa Padua, Timothy T Rogers

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、cs.LG

2308.11521 2023-08-28 cs.CL cs.AI cs.LG 90%

Self-Deception: Reverse Penetrating the Semantic Firewall of Large Language Models

Zhenhua Wang, Wei Xie, Kai Chen, Baosheng Wang, Zhiwen Gui, Enze Wang

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、cs.LG

Comments Serious errors were found in the experiment, which may lead to the overturning of the overall conclusions of the paper

2307.16139 2023-08-01 cs.CL cs.AI cs.LG 90%

User-Controlled Knowledge Fusion in Large Language Models: Balancing Creativity and Hallucination

Chen Zhang

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、cs.LG

2208.11057 2023-06-21 cs.CL cs.AI 90%

Prompting as Probing: Using Language Models for Knowledge Base Construction

Dimitrios Alivanistos, Selene Báez Santamaría, Michael Cochez, Jan-Christoph Kalo, Emile van Krieken, Thiviyan Thanapalasingam

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

Comments Published in LM-KBC 22: Knowledge Base Construction from Pre-trained Language Models, Challenge at ISWC 2022. 12+12 pages

2608.19579 2026-08-21 cs.AI math.DS 新提交 90%

Enforcing LLM Safety through DMD-based Classification of Prompt-Response Embedding Dynamics

基于DMD的提示-响应嵌入动态分类实现大语言模型安全

Mohamed Akrout, Olivera Kotevska, Dan Wilson

机构 * University of Tennessee（田纳西大学）； Oak Ridge National Laboratory（橡树岭国家实验室）

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

AI总结本文将幻觉检测的动力系统框架扩展到LLM安全分类，通过拟合安全与不安全状态的Koopman模型，利用差分残差评分分类不安全输出，在多基准测试中验证了纳入提示嵌入的改进效果。

2608.14352 2026-08-17 cs.SE cs.LG 新提交 90%

ATLAS: Discovering Agent Strategies through LLM-Guided Abstraction and Automata Learning

ATLAS：通过大语言模型引导的抽象与自动机学习发现智能体策略

Ignacio D. Lopez-Miguel, Andreas Happe, Jürgen Cito, Ezio Bartocci, Bettina Könighofer, Martin Tappler

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

AI总结该研究提出ATLAS方法，结合轨迹抽象与自动机学习从智能体轨迹推断有限状态模型，以实现对LLM智能体行为的可解释分析、知识迁移及简洁解释。

Comments 7 pages, accepted for publication at ACM/IEEE MODELS 2026

2608.08127 2026-08-11 cs.AI cs.LO cs.SE 新提交 90%

Improving Constraint Models with LLM Agents

用大语言模型智能体改进约束模型

Florentina Voboril, Stefan Szeider

机构 * TU Wien（维也纳技术大学）

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

AI总结该研究提出一种基于LLM智能体的框架，通过迭代诊断修复改进约束模型，在9个组合优化问题的27个测试实例中多数表现优于原模型，部分问题求解速度提升超两个数量级，证明自主智能体方法可支持约束模型优化。

2606.00898 2026-08-11 cs.CL cs.DL 版本更新 90%

Citation Grounding Measures the Oracle: Graph Coverage Determines Reported LLM Hallucination Rates in Law

引用溯源：通过法律引用图检测和减少LLM引用幻觉

Volodymyr Ovcharov

机构 * LEX AI LLC

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

AI总结提出引用溯源（CG）指标，利用乌克兰法院判决的引用图（1.008亿判决，5.02亿边）检测LLM法律引用幻觉，并通过CG-DPO方法（基于真实判决构建偏好对）减少幻觉，在100个法律查询上CG为0.791-0.873，幻觉率13-21%。

Comments 21 pages, 4 figures, 5 tables. Substantially revised: title, framing and several v1 results changed. Adds a coverage sweep and a separability analysis; corrects the DPO configuration, the density-accuracy correlation and the qualitative examples. Code and data: https://huggingface.co/datasets/overthelex/citation-grounding-eval

2604.25921 2026-08-11 cs.CL cs.CR 版本更新 90%

One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety

逐词进行：增量完成分解打破LLM安全

Samee Arif, Naihao Deng, Zhijing Jin, Rada Mihalcea

机构 * University of Michigan（密歇根大学）； University of Toronto（多伦多大学）

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

AI总结本文提出增量完成分解（ICD）策略，通过逐词生成恶意请求相关词来突破LLM安全机制，评估多种变体在多个基准测试中表现优异，并理论解释其有效性。

2607.24072 2026-07-28 cs.CL 新提交 90%

LLM-Based vs. Lexicon-Based Sentiment Signals for Tail-Risk Detection in Meme Stocks

基于大语言模型与基于词典的情绪信号在表情包股票尾部风险检测中的比较

Paul Kilian, Markus Kleffmann

机构 * IU International University of Applied Sciences（IU国际应用科学大学）

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

AI总结比较基于词典和基于大语言模型的情绪分析在表情包股票尾部风险检测中的应用，构建情绪指标并评估与市场回报关系，结果显示基于LLM的指标表征更丰富，但与市场走势关系因资产而异。

Comments 10 pages, 1 figure, 4 tables

2602.11619 2026-07-16 cs.AI 版本更新 90%

When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents

当智能体与自身意见相左：测量基于LLM的智能体的行为一致性

Aman Mehta

机构 * Aman Mehta

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

AI总结研究发现基于LLM的智能体在相同任务上运行结果不一致，且这种不一致与任务成功率密切相关，通过监控行为一致性可提升智能体可靠性。

Comments Accepted at the ICML 2026 Workshop on Statistical Frameworks for Uncertainty in Agentic Systems. 12 pages, 9 figures

2607.03350 2026-07-07 cs.CR cs.AI cs.SE 新提交 90%

LLM-Enhanced Hierarchical Heterogeneous Graph Representation Learning for Malicious Python Package Detection

用于恶意Python包检测的基于大语言模型增强的分层异构图表示学习

Hang Gao, Xiaoyu Chen, Baoquan Cui, Zhen Tang, Peng Qiao, Fengge Wu, Jian Zhang

机构 * Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）； University of Chinese Academy of Sciences（中国科学院大学）

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

AI总结提出用于恶意Python包检测的LLM增强分层异构图表示学习框架，构建分层异构代码图，利用LLMs推理功能语义角色，开发分层异构图神经网络并结合功能级归因机制，实验表明该框架性能优越。

2606.30689 2026-07-01 cs.SE cs.AI 新提交 90%

Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code

规范驱动开发中的引用规范：LLM生成代码的输出确定性与自动幻觉检测的跨模型实证研究

Subham Panda

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

AI总结通过对比三种规范驱动开发框架，发现强制引用要求会降低输出确定性但能实现自动幻觉检测，该权衡在不同模型间一致。

Comments 17 pages

2606.28798 2026-06-30 cs.AI stat.AP 90%

Primary ICD Category Prediction using LLM-based Probing

基于LLM探针的主要ICD类别预测

Chengyuan Liu, Xinyue Zhang, Yao Li, Guanting Chen

机构 * Department of Statistics, Pennsylvania State University（宾夕法尼亚州立大学统计学系）； Department of Biostatistics, University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校生物统计学系）； Department of Statistics and Operations Research, University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校统计学与运筹学系）

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

AI总结本研究利用冻结的医学大语言模型表示作为共享嵌入空间，通过线性探针融合结构化变量和临床叙述，实现多模态主要诊断类别预测，在MIMIC-IV上达到87.69%的严格准确率。

Comments 9 pages, 2 figures. Supplementary materials provided as an ancillary file

2606.22741 2026-06-23 cs.LG 新提交 90%

GRADE: Graph Representation of LLM Agent Dependency and Execution

GRADE: LLM智能体依赖与执行的图表示

Yue Zhao

机构 * University of Southern California（南加州大学）

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

AI总结提出GRADE，将LLM智能体的运行建模为具有执行边和依赖边的图，依赖边通过分级推断，在多个数据集上优于运行规模指标，并可用于故障定位。

Comments 18 pages, 5 figures, 8 tables. Code: https://github.com/yzhao062/grade

2606.21399 2026-06-23 cs.AI 新提交 90%

Calibration Is Not Control: Why LLM-Agent Oversight Needs Intervention

校准不是控制：为什么LLM代理监督需要干预

Chubin Zhang, Zhenglin Wan, Xingrui Yu, Jingxuan Wu, Qi Wen, Pengfei Zhou, Wangbo Zhao, Ivor Tsang

机构 * Nanyang Technological University（南洋理工大学）； National University of Singapore（新加坡国立大学）； CFAR Agency for Science Technology and Research（科技研究局CFAR）； IHPC Agency for Science Technology and Research（科技研究局IHPC）； Department of Statistics and Operations Research UNC-Chapel Hill（北卡罗来纳大学教堂山分校统计与运筹学系）

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

AI总结本文指出LLM代理运行时监督中常用的标量风险预测存在目标错误，提出以干预优势为决策对象，并引入前缀分支方法进行动作条件控制，实验表明该方法能显著降低控制遗憾。

Comments 29 pages

1. 知识编辑与模型理解 7552 篇

Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Maps

TracrBench: Generating Interpretability Testbeds with Large Language Models

Towards Signal Processing In Large Language Models

Ovis: Structural Embedding Alignment for Multimodal Large Language Model

Fact-Checking the Output of Large Language Models via Token-Level Uncertainty Quantification

SPIN: Sparsifying and Integrating Internal Neurons in Large Language Models for Text Classification

Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models

Detoxifying Large Language Models via Knowledge Editing

InstructEdit: Instruction-based Knowledge Editing for Large Language Models

LLMCheckup: Conversational Examination of Large Language Models via Interpretability Tools and Self-Explanations

Unified Scene Representation and Reconstruction for 3D Large Language Models

DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

Rethinking Interpretability in the Era of Large Language Models

Explainability for Large Language Models: A Survey

Conceptual structure coheres in human cognition but not in large language models

Self-Deception: Reverse Penetrating the Semantic Firewall of Large Language Models

User-Controlled Knowledge Fusion in Large Language Models: Balancing Creativity and Hallucination

Prompting as Probing: Using Language Models for Knowledge Base Construction

Enforcing LLM Safety through DMD-based Classification of Prompt-Response Embedding Dynamics

ATLAS: Discovering Agent Strategies through LLM-Guided Abstraction and Automata Learning

Improving Constraint Models with LLM Agents

Citation Grounding Measures the Oracle: Graph Coverage Determines Reported LLM Hallucination Rates in Law

One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety

LLM-Based vs. Lexicon-Based Sentiment Signals for Tail-Risk Detection in Meme Stocks

When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents

LLM-Enhanced Hierarchical Heterogeneous Graph Representation Learning for Malicious Python Package Detection

Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code

Primary ICD Category Prediction using LLM-based Probing

GRADE: Graph Representation of LLM Agent Dependency and Execution

Calibration Is Not Control: Why LLM-Agent Oversight Needs Intervention