语言大模型 / LLM - arXivDaily 专题

2310.06450 2023-10-12 cs.CL cs.AI 90%

Constructive Large Language Models Alignment with Diverse Feedback

Tianshu Yu, Ting-En Lin, Yuchuan Wu, Min Yang, Fei Huang, Yongbin Li

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

2310.00789 2023-10-03 cs.CL cs.LG 90%

Testing the Limits of Unified Sequence to Sequence LLM Pretraining on Diverse Table Data Tasks

Soumajyoti Sarkar, Leonard Lausen

专题命中预训练与数据：pretraining(title,abstract)；LLM(title)；large language model(abstract)；language model(abstract)

2309.10707 2023-09-20 eess.AS cs.CL cs.LG cs.SD 90%

Corpus Synthesis for Zero-shot ASR domain Adaptation using Large Language Models

Hsuan Su, Ting-Yao Hu, Hema Swetha Koppula, Raviteja Vemulapalli, Jen-Hao Rick Chang, Karren Yang, Gautam Varma Mantena, Oncel Tuzel

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2308.14242 2023-08-29 cs.AI cs.CL 90%

The Cultural Psychology of Large Language Models: Is ChatGPT a Holistic or Analytic Thinker?

Chuanyang Jin, Songyang Zhang, Tianmin Shu, Zhihan Cui

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

2308.10502 2023-08-22 cs.LG cs.CL stat.ML 90%

GradientCoin: A Peer-to-Peer Decentralized Large Language Models

Yeqi Gao, Zhao Song, Junze Yin

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2306.15766 2023-06-29 cs.CL cs.LG 90%

Large Language Models as Annotators: Enhancing Generalization of NLP Models at Minimal Cost

Parikshit Bansal, Amit Sharma

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2306.12925 2023-06-23 cs.CL cs.AI cs.SD eess.AS stat.ML 90%

AudioPaLM: A Large Language Model That Can Speak and Listen

Paul K. Rubenstein, Chulayuth Asawaroengchai, Duc Dung Nguyen, Ankur Bapna, Zalán Borsos, Félix de Chaumont Quitry, Peter Chen, Dalia El Badawy, Wei Han, Eugene Kharitonov, Hannah Muckenhirn, Dirk Padfield, James Qin, Danny Rozenberg, Tara Sainath, Johan Schalkwyk, Matt Sharifi, Michelle Tadmor Ramanovich, Marco Tagliasacchi, Alexandru Tudor, Mihajlo Velimirović, Damien Vincent, Jiahui Yu, Yongqiang Wang, Vicky Zayats, Neil Zeghidour, Yu Zhang, Zhishuai Zhang, Lukas Zilka, Christian Frank

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；pretraining(abstract)；分类 cs.CL、cs.AI

Comments Technical report

2305.07922 2023-05-23 cs.CL cs.LG cs.PL 90%

CodeT5+: Open Code Large Language Models for Code Understanding and Generation

Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi D. Q. Bui, Junnan Li, Steven C. H. Hoi

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；pretraining(abstract)；分类 cs.CL、cs.LG

Comments 26 pages, preprint

2302.07388 2023-02-16 cs.CL cs.AI 90%

Adding Instructions during Pretraining: Effective Way of Controlling Toxicity in Language Models

Shrimai Prabhumoye, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro

专题命中预训练与数据：language model(title,abstract)；pretraining(title,abstract)；large language model(abstract)；分类 cs.CL、cs.AI

Comments This paper will be presented at EACL 2023

2209.10063 2023-01-26 cs.CL cs.AI 90%

Generate rather than Retrieve: Large Language Models are Strong Context Generators

Wenhao Yu, Dan Iter, Shuohang Wang, Yichong Xu, Mingxuan Ju, Soumya Sanyal, Chenguang Zhu, Michael Zeng, Meng Jiang

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；prompting(abstract)；分类 cs.CL、cs.AI

Comments Accepted at ICLR 2023 (v3, add code and implementation details)

2209.15003 2022-10-03 cs.CL cs.AI 90%

Compositional Semantic Parsing with Large Language Models

Andrew Drozdov, Nathanael Schärli, Ekin Akyürek, Nathan Scales, Xinying Song, Xinyun Chen, Olivier Bousquet, Denny Zhou

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；prompting(abstract)；分类 cs.CL、cs.AI

Comments Fixed metadata. No other changes

2205.13621 2022-09-12 cs.CL cs.LG 90%

Differentially Private Decoding in Large Language Models

Jimit Majmudar, Christophe Dupuy, Charith Peris, Sami Smaili, Rahul Gupta, Richard Zemel

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

2205.12600 2022-05-26 cs.CL cs.LG 90%

ORCA: Interpreting Prompted Language Models via Locating Supporting Data Evidence in the Ocean of Pretraining Data

Xiaochuang Han, Yulia Tsvetkov

专题命中预训练与数据：language model(title,abstract)；pretraining(title,abstract)；prompting(abstract)；分类 cs.CL、cs.LG

2107.06955 2021-07-16 cs.CL cs.LG 90%

HTLM: Hyper-Text Pre-Training and Prompting of Language Models

Armen Aghajanyan, Dmytro Okhonko, Mike Lewis, Mandar Joshi, Hu Xu, Gargi Ghosh, Luke Zettlemoyer

专题命中预训练与数据：language model(title,abstract)；prompting(title,abstract)；pretraining(abstract)；分类 cs.CL、cs.LG

2002.02000 2020-02-07 cs.CL cs.LG 90%

Aligning the Pretraining and Finetuning Objectives of Language Models

Nuo Wang Pierse, Jingwen Lu

专题命中预训练与数据：language model(title,abstract)；pretraining(title,abstract)；small language model(abstract)；分类 cs.CL、cs.LG

Comments 8 pages, 2 figures and 6 tables. Submitted to ICML 2020

2506.13396 2025-07-08 cs.CL eess.AS 90%

Bi-directional Context-Enhanced Speech Large Language Models for Multilingual Conversational ASR

Yizhou Peng, Hexin Liu, Eng Siong Chng

机构 * Alibaba-NTU Global e-Sustainability CorpLab（阿里巴巴-国立科技大学全球可持续发展公司实验室）； Nanyang Technological University（国立科技大学）； College of Computing and Data Science（计算与数据科学学院）

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；SLM(abstract,comments)；分类 cs.CL

Comments Accepted By Interspeech 2025 MLC-SLM workshop as a Research Paper

2502.12150 2025-06-17 cs.CL 90%

Idiosyncrasies in Large Language Models

Mingjie Sun, Yida Yin, Zhiqiu Xu, J. Zico Kolter, Zhuang Liu

机构 * Carnegie Mellon University（卡内基梅隆大学）； Princeton University（普林斯顿大学）； University of Pennsylvania（宾夕法尼亚大学）

专题命中预训练与数据：large language model(title,abstract)；language model(title,abstract)；LLM(abstract,comments)；分类 cs.CL

Comments Published in ICML 2025. Website at https://eric-mingjie.github.io/llm-idiosyncrasies/index.html

2608.13545 2026-08-14 cs.CL cs.AI cs.LG 新提交 89%

LittleLearner: Language Models Under Pedagogically Controlled Knowledge Exposure

LittleLearner：受教学可控知识暴露约束的语言模型

Fanfei Li, Jana Zeller, Manuel Prada-Corral, Thaddäus Wiedemer, Prasanna Mayilvahanan, Ryan Cotterell, Wieland Brendel

机构 * Ellis Institute（埃利斯研究所）

专题命中预训练与数据：language model(title,abstract)；LLM(abstract,abstract_cn)；pretraining(abstract)；post-training(abstract)

AI总结本研究推出专为美国小学定制的LITTLECURRICULUM语料库，训练得到受知识边界约束的LittleLearner模型，构建沙盒用于研究模型知识习得，实验证实其注入新知识后不会提升超范围能力。

2608.12741 2026-08-14 cs.IR 新提交 89%

Knowledge Synthesis Review Framework: Task-Level Benchmarking of LLM-Based Systems for Multi-Source Evidence Synthesis

知识综合评审框架：面向多源证据综合的基于大语言模型系统的任务级基准测试

Wafa Shafqat, Mark Patterson, Steven N. Liss

专题命中预训练与数据：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结本研究提出KSR人机协作框架，将证据综合拆分为4项任务，对4款LLM系统开展任务级基准测试，发现各系统各有优劣，该框架可揭示单源综合遗漏的信息，且透明可审计。

2608.09567 2026-08-11 cs.CR 新提交 89%

From Runnable to Verifiable: An Independent Reproducibility Study of LLM/Agent-Driven Vulnerability Validation Artifacts

从可运行到可验证：LLM/智能体驱动的漏洞验证制品的独立可复现性研究

Bo Chen

专题命中预训练与数据：LLM(title,title_cn)

AI总结本研究通过预注册可复现性审计，发现LLM/智能体驱动的漏洞验证制品存在标识符不一致、可运行率低、预言机不可靠等问题，提出的审计方案可作为安全可复现性社区的复用模板。

Comments 16 pages, 2 figures, 6 tables

2608.08413 2026-08-11 cs.SE 新提交 89%

Tangent: An Empirical Study of Testing Practices for LLM-Based Agent Applications

Tangent：基于大语言模型的智能体应用测试实践的实证研究

Rangeet Pan, Tyler Stennett, Divya Sankar, Bridget McGinn, Alessandro Orso, Raju Pavuluri, Saurabh Sinha, Maja Vukovic

专题命中预训练与数据：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结 Tangent通过对开源项目和行业从业者的研究，分析了LLM智能体应用的测试现状，发现其以单元测试为主、存在覆盖不足等问题，并提出了相关研究方向。

Comments Accepted at ASE'26

2509.25465 2026-08-06 cs.SE 版本更新 89%

BloomAPR: A Bloom's Taxonomy-based Framework for Assessing the Capabilities of LLM-Powered APR Solutions

BloomAPR：基于布鲁姆教育目标分类学的框架，用于评估大语言模型驱动的自动程序修复（APR）方案的能力

Yinghang Ma, Jiho Shin, Leuson Da Silva, Zhen Ming, Jiang, Song Wang, Foutse Khomh, Shin Hwei Tan

专题命中预训练与数据：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结 BloomAPR是基于布鲁姆教育目标分类学的动态评估框架，评估了ChatRepair、CigaR等APR方案在不同LLM及布鲁姆分类层级下的漏洞修复能力，发现其存在性能差异并指出基准演进的必要性。

Comments 22 pages, 7 figures, Manuscript submitted to ACM Transactions on Software Engineering and Methodology

2608.00661 2026-08-04 cs.SE 新提交 89%

Unreliable in Practice? A Comprehensive Study of Errors in LLM-Generated Code

实践中不可靠？对大语言模型生成代码中的错误的综合研究

Rodrigo Pato Nogueira, Marco Vieira, João R. Campos

专题命中预训练与数据：LLM(title,summary_cn)；large language model(abstract)；language model(abstract)

AI总结本研究分析7款LLMs生成的86726个含错代码样本，对比不同模型、语言的错误模式，发现大型模型也常犯简单错误，代码常省略关键安全检查，为提升LLM生成代码可靠性提供依据。

Comments Accepted for publication in the 37th IEEE International Symposium on Software Reliability Engineering (ISSRE) 2026

2607.00018 2026-07-02 cs.CY 新提交 89%

The Limits of LLM Forecasting: Parametric Knowledge Gaps Across Conflict Zones

LLM预测的局限性：冲突区域间的参数化知识差距

Poli Nemkova

专题命中预训练与数据：LLM(title,title_cn)

AI总结研究发现LLM在冲突预测中无法区分稳定与升级期，对低覆盖地区预测为总是升级，对高覆盖地区预测为从不升级，性能低于简单逻辑回归模型。

2606.29632 2026-06-30 eess.AS cs.CV cs.SD 89%

VIB-AVSR: Variational Information Bottleneck for Noise-Robust LLM-Based Audio-Visual Speech Recognition

VIB-AVSR：基于变分信息瓶颈的噪声鲁棒LLM视听语音识别

Piyush Arora, Navlika Singh, Umberto Cappellazzo, Stavros Petridis, Maja Pantic

机构 * Imperial College London（帝国理工学院伦敦分校）； NatWest AI Research（NatWest人工智能研究）

专题命中预训练与数据：LLM(title,title_cn)

AI总结提出VIB-AVSR，通过在LLM骨干中插入变分信息瓶颈层来正则化表示，无需架构修改或额外数据，即可在多种噪声条件下提升AVSR鲁棒性。

Comments Accepted to INTERSPEECH 2026. Our code is available at https://github.com/PiyushArora1010/VIB-AVSR

2606.23911 2026-06-24 cs.IR 新提交 89%

Scaling Dense Retrieval with LLM-Annotated Training Data: Structured Mining and Progressive Curriculum for E-Commerce Sponsored Search

使用LLM标注的训练数据扩展稠密检索：面向电商赞助搜索的结构化挖掘与渐进式课程

Md Omar Faruk Rokon, Shasvat Desai, Jhalak Nilesh Acharya, Isha Shah, Kumar Priyam, Brahanyaa Somasundaram, Vamsee Tangirala, Minuteresa Thomas, Vivek Arora, Vijay Manchi, Hong Yao, Kuang-chih Lee

专题命中预训练与数据：LLM(title,title_cn)

AI总结针对电商搜索中点击信号偏差和人工标注成本高的问题，提出利用多通道检索挖掘、校准的LLM级联标注和五级渐进式课程训练，在Walmart搜索中实现NDCG@10提升5.1%，尾查询提升显著。

Comments Accepted at E-Commerce Workshop, SIGIR 2026

2606.01475 2026-06-02 cs.CY 89%

An LLM-based Chain-of-Response Counter-Scam System

基于LLM的链式响应反诈骗系统

Heedou Kim, Mogan Gim, Donghee Choi, Hoonick Lee, Soonil Bae, Mi-Young Kim, Jaewoo Kang

专题命中预训练与数据：LLM(title,title_cn)

AI总结提出Counter Scam框架，利用LLM多智能体协同实现从检测到调查的端到端反诈骗响应，通过CSRA、CSRT和CSRD组件提升效率，实验表明微调sLLM在CSRT任务上超越商业模型10%以上。

Comments This paper has been accepted for publication at IJCAI 2026

2605.26437 2026-05-27 econ.GN q-fin.EC 89%

Divergent Minds, Convergent Baselines: A Bounded-Rationality Account of LLM-Human Strategic Behaviour

分歧的思维，趋同的基线：LLM与人类战略行为的有界理性解释

Po Han Teo

专题命中预训练与数据：LLM(title,title_cn)

AI总结本文提出有界理性框架，将人类与LLM在战略博弈中的行为差异归因于计算约束的不同，并给出四个操作测试来区分两者的偏差项δ。

Comments 12 pages, 1 table, no figures. Theoretical prequel paper

2605.22005 2026-05-26 cs.LG cs.AI cs.CL 89%

Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)

检查你的大语言模型的秘密词典！五行代码揭示你的大语言模型学到了什么（包括它不应该学到的）

Hisashi Miyashita

机构 * Mgnite Inc.（Mgnite公司）

专题命中预训练与数据：LLM(title,abstract)；large language model(abstract)；language model(abstract)；pretraining(abstract)

AI总结通过对lm_head权重矩阵进行奇异值分解（仅需五行PyTorch代码且无需模型推理），直接从模型权重中揭示可解释的语义子空间，并发现模型训练数据组成和策展哲学。

2603.26089 2026-05-12 cs.LG cs.AI cs.CL 89%

Selective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind

对LLM心理自我建模能力的有选择性缺陷的测试：基于行为的理论自我认知测试

Christopher Ackerman

专题命中预训练与数据：LLM(title,title_cn)；分类 cs.CL、cs.AI、cs.LG

AI总结本文通过行为测试探讨LLM在理论自我认知任务中的能力，发现较早的LLM无法完成任务，而较新的LLM在他人认知建模上表现接近人类，但自我建模仍存在缺陷，除非使用推理轨迹辅助。

Comments 22 pages, 13 figures, 1 table

1. 预训练与数据 12318 篇

Constructive Large Language Models Alignment with Diverse Feedback

Testing the Limits of Unified Sequence to Sequence LLM Pretraining on Diverse Table Data Tasks

Corpus Synthesis for Zero-shot ASR domain Adaptation using Large Language Models

The Cultural Psychology of Large Language Models: Is ChatGPT a Holistic or Analytic Thinker?

GradientCoin: A Peer-to-Peer Decentralized Large Language Models

Large Language Models as Annotators: Enhancing Generalization of NLP Models at Minimal Cost

AudioPaLM: A Large Language Model That Can Speak and Listen

CodeT5+: Open Code Large Language Models for Code Understanding and Generation

Adding Instructions during Pretraining: Effective Way of Controlling Toxicity in Language Models

Generate rather than Retrieve: Large Language Models are Strong Context Generators

Compositional Semantic Parsing with Large Language Models

Differentially Private Decoding in Large Language Models

ORCA: Interpreting Prompted Language Models via Locating Supporting Data Evidence in the Ocean of Pretraining Data

HTLM: Hyper-Text Pre-Training and Prompting of Language Models

Aligning the Pretraining and Finetuning Objectives of Language Models

Bi-directional Context-Enhanced Speech Large Language Models for Multilingual Conversational ASR

Idiosyncrasies in Large Language Models

LittleLearner: Language Models Under Pedagogically Controlled Knowledge Exposure

Knowledge Synthesis Review Framework: Task-Level Benchmarking of LLM-Based Systems for Multi-Source Evidence Synthesis

From Runnable to Verifiable: An Independent Reproducibility Study of LLM/Agent-Driven Vulnerability Validation Artifacts

Tangent: An Empirical Study of Testing Practices for LLM-Based Agent Applications

BloomAPR: A Bloom's Taxonomy-based Framework for Assessing the Capabilities of LLM-Powered APR Solutions

Unreliable in Practice? A Comprehensive Study of Errors in LLM-Generated Code

The Limits of LLM Forecasting: Parametric Knowledge Gaps Across Conflict Zones

VIB-AVSR: Variational Information Bottleneck for Noise-Robust LLM-Based Audio-Visual Speech Recognition

Scaling Dense Retrieval with LLM-Annotated Training Data: Structured Mining and Progressive Curriculum for E-Commerce Sponsored Search

An LLM-based Chain-of-Response Counter-Scam System

Divergent Minds, Convergent Baselines: A Bounded-Rationality Account of LLM-Human Strategic Behaviour

Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)

Selective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind