arXivDaily arXiv每日学术速递 周一至周五更新

AI 大模型

大模型推理能力

大模型数学、逻辑、规划、多步推理和测试时计算能力。

共收录 10507 信号源:cs.CL, cs.AI, cs.LG

1. 推理评测 10507 篇

2109.04912 2021-09-13 cs.CL cs.AI cs.LG 67%

ReasonBERT: Pre-trained to Reason with Distant Supervision

Xiang Deng, Yu Su, Alyssa Lees, You Wu, Cong Yu, Huan Sun

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments Accepted to EMNLP'2021. Our code and pre-trained models are available at https://github.com/sunlab-osu/ReasonBERT

详情

展开后加载摘要…

URL PDF HTML 收藏
2109.04312 2021-09-10 cs.CL cs.AI cs.IR cs.LG 67%

MATE: Multi-view Attention for Table Transformer Efficiency

Julian Martin Eisenschlos, Maharshi Gor, Thomas Müller, William W. Cohen

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments Accepted to EMNLP 2021

详情

展开后加载摘要…

URL PDF HTML 收藏
2011.04767 2020-11-16 cs.CL cs.AI cs.LG 67%

An Analysis of Dataset Overlap on Winograd-Style Tasks

Ali Emami, Adam Trischler, Kaheer Suleman, Jackie Chi Kit Cheung

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments 11 pages with references, accepted at COLING 2020

Journal ref Coling2020

详情

展开后加载摘要…

URL PDF HTML 收藏
2011.04006 2020-11-10 cs.LG cs.AI cs.CL cs.CV cs.IR 67%

Long Range Arena: A Benchmark for Efficient Transformers

Yi Tay, Mostafa Dehghani, Samira Abnar, Yikang Shen, Dara Bahri, Philip Pham, Jinfeng Rao, Liu Yang, Sebastian Ruder, Donald Metzler

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

详情

展开后加载摘要…

URL PDF HTML 收藏
2004.14507 2020-10-12 cs.LG cs.AI cs.CL 67%

Counterfactual Off-Policy Training for Neural Response Generation

Qingfu Zhu, Weinan Zhang, Ting Liu, William Yang Wang

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

详情

展开后加载摘要…

URL PDF HTML 收藏
2007.10866 2020-07-22 cs.CL cs.AI cs.LG 67%

IITK-RSA at SemEval-2020 Task 5: Detecting Counterfactuals

Anirudh Anil Ojha, Rohin Garg, Shashank Gupta, Ashutosh Modi

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments 10 pages, 1 figure, 4 tables. For associated code, see https://github.com/gargrohin/Counterfactuals-NLP. Accepted at Proceedings of 14th International Workshop on Semantic Evaluation (SemEval-2020)

详情

展开后加载摘要…

URL PDF HTML 收藏
2005.06249 2020-05-14 cs.CL cs.AI cs.LG 67%

Machine Reading Comprehension: The Role of Contextualized Language Models and Beyond

Zhuosheng Zhang, Hai Zhao, Rui Wang

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments 51 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
1907.10738 2019-07-26 cs.CL cs.AI cs.IR cs.LG 67%

Careful Selection of Knowledge to solve Open Book Question Answering

Pratyay Banerjee, Kuntal Kumar Pal, Arindam Mitra, Chitta Baral

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI、cs.LG

Comments Accepted to ACL 2019

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.05183 2026-08-18 cs.CL cs.AI cs.HC 版本更新 66%

The Granularity Gap: A Multi-Dimensional Cross-Generational Audit of Sycophancy in Gemini Models

粒度差距:Gemini 模型中谄媚行为的多维纵向审计

Patrick Keough

机构 * Independent Researcher(独立研究者)

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.CL、cs.AI

AI总结 通过多维度分级评估(Likert 0-4),揭示 Gemini 模型在连续尺度上的谄媚行为,发现粗粒度二值指标掩盖了大量社会顺从行为,且代际进步非单调,存在对齐税(谄媚与真实性负相关)。

Comments v2: Major correction. Three v1 claims withdrawn (U-shaped detection curve, recalibration remedy, one reliability figure); the central 29% result survives. Adds a four-judge panel over a stratified 1,200-response sample, 10,792 votes with written reasoning. Data unchanged from v1. 21 pages, 8 figures, 18 tables. Itemized changelog and code: https://github.com/pskeough/The-Granularity-Gap

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.08442 2026-08-05 cs.CR cs.AI cs.LG 版本更新 66%

Injection-Execution Dissociation: A Mechanistic Evaluation of Persistent Memory Attacks and Defenses in Stateful LLM Agents

多层架构中的防御效果:对持久内存攻击状态机LLM代理的机制性评估

Jun Wen Leong

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.AI、cs.LG

AI总结 本文评估了六种防御措施在九个开源模型上的延迟触发攻击效果,发现输入级和检索级过滤器效果不佳,而内存层工具门控(Memory Sandbox)显著降低攻击成功率,揭示了不同防御类别的失效原因。

Comments v4: Added double dissociation (reasoning-mode ablation), content-layer defense (RATG), loaded-corpus frontier evaluation (21 models, 3 providers, N=40), 7B judge capability bound, reproducibility validity criterion, ethics/disclosure statement. Gemini 3.1 Pro Preview 95% ASR; GPT-5.1 regression (22.5%); tripartite vendor divergence. Code: github.com/junwenleong/stateful-agent-security-eval

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.13070 2025-12-16 cs.AI cs.CL 66%

M-GRPO: Stabilizing Self-Supervised Reinforcement Learning for Large Language Models with Momentum-Anchored Policy Optimization

M-GRPO:通过动量锚定策略优化稳定大语言模型的自监督强化学习

Bizhe Bai, Hongming Wu, Peng Ye, Tao Chen

机构 * Shanghai Innovation Institute(上海创新研究院) College of Future Information Technology, Fudan(复旦大学未来信息技术学院) Shanghai AI Laboratory(上海人工智能实验室) The Chinese University of Hong Kong(香港中文大学)

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.CL、cs.AI

AI总结 M-GRPO通过动量锚定策略优化和IQR过滤方法,稳定大语言模型的自监督强化学习训练,提升训练稳定性和性能。

Comments 7 pages, 5 figures,Accepted NeurIPS 2025 Workshop on Efficient Reasoning

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.23083 2025-12-11 cs.AI cs.LG cs.SE 66%

Smaller Models, Smarter Rewards: A Two-Sided Approach to Process and Outcome Rewards

更小的模型,更聪明的奖励:一种双面方法来处理和结果奖励

Jan Niklas Groeneveld, Xi Qin, Alexander Schaefer, Yaad Oren

机构 * University of California, Irvine(加州大学尔湾分校) SAP Lab(SAP实验室) Stanford Human-Centered AI Institution(斯坦福人本AI机构)

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.AI、cs.LG

AI总结 本文提出了一种双面方法,利用小型语言模型生成高质量代码,通过融合过程和结果奖励,提升了代码生成的搜索能力。

Comments Accepted and presented at NeurIPS 2025 Workshop: Foundations of Reasoning in Language Models

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.14705 2025-01-27 cs.LG cs.CL 66%

The Karp Dataset

Mason DiCicco, Eamon Worden, Conner Olsen, Nikhil Gangaram, Daniel Reichman, Neil Heffernan

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.CL、cs.LG

Comments Accepted to the 4th workshop on mathematical reasoning and AI at NeurIPS 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2406.09308 2024-06-14 cs.CL cs.LG 66%

Transformers meet Neural Algorithmic Reasoners

Wilfried Bounsi, Borja Ibarz, Andrew Dudzik, Jessica B. Hamrick, Larisa Markeeva, Alex Vitvitskyi, Razvan Pascanu, Petar Veličković

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.CL、cs.LG

Comments To appear at CVPR 2024 Multimodal Algorithmic Reasoning (MAR) Workshop. 10 pages, 5 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
1905.06393 2019-05-17 cs.LG cs.AI stat.ML 66%

IPC: A Benchmark Data Set for Learning with Graph-Structured Data

Patrick Ferber, Tengfei Ma, Siyu Huo, Jie Chen, Michael Katz

专题命中 推理评测 :planning(abstract);分类 cs.AI、cs.LG;reasoning(comments)

Comments ICML 2019 Workshop on Learning and Reasoning with Graph-Structured Data. The data set is accessible from https://github.com/IBM/IPC-graph-data

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.06802 2026-06-24 cs.AI 版本更新 65%

Riemann-Bench: A Benchmark for Moonshot Mathematics

Riemann-Bench: 面向登月级数学的基准测试

Suhaas Garre, Erik Knutsen, Sushant Mehta, Edwin Chen

机构 * Surge AI

专题命中 推理评测 :reasoning(abstract,comments);分类 cs.AI;logical reasoning(comments)

AI总结 提出Riemann-Bench基准,由专家设计研究级数学问题,评估AI系统超越奥数水平的推理能力,结果显示前沿模型得分低于10%。

Comments Accepted to Logical Reasoning of Large Language Models, ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.21928 2026-08-25 cs.AI cs.CL cs.RO 新提交 62%

GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI

GuardianBench:面向具身智能中潜在上下文风险的同场景指令对比基准

Zhesheng Zhang, Jiahao Lu, Wei Liu, Cong Pan, Jianhua Yang, Yixiang Chen, Hongyuan Yu, Mengqi Zhang, Kailin Lyu, Zhumin Chen, Keji He

机构 * Shandong University(山东大学) National University of Singapore(新加坡国立大学) Nanjing University of Aeronautics and Astronautics(南京航空航天大学) Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所) Xiaomi Corporation(小米公司)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 该研究提出基于国际安全标准的同场景指令对比基准GuardianBench,发现VLMs对指令不敏感,用轻量级目标VLOS可提升其安全推理性能。

Comments 21 pages, 4 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.21409 2026-08-25 cs.CY cs.AI cs.CL 新提交 62%

Sycophants in the Courtroom: Are LLMs Fragile to Juridical Authority and Evolving Legal Standards?

法庭中的谄媚者:大型语言模型(LLMs)是否对司法权威与演变的法律标准脆弱?

Lorenzo Molfetta, Alessio Cocchieri, Luca Ragazzi, Ilaria Bartolini, Marco Patella, Gianluca Moro

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 该研究通过比较诊断框架对比LLMs在法律与医学领域的表现,发现法律LLMs对司法权威扰动更脆弱,过度信任权威虚假信息,模型规模会放大该问题。

Comments Please cite the definitive, peer-reviewed version of this article published in the Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), edited by Maria Liakata et al., Association for Computational Linguistics, pp. 10865-10886, 2026. DOI: https://doi.org/10.18653/v1/2026.acl-long.497

Journal ref Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), San Diego, California, United States, Association for Computational Linguistics, 2026, pp. 10865-10886

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.15382 2026-08-25 cs.AI cs.CL cs.IR q-bio.QM 版本更新 62%

Framework for Grounding Healthcare LLMs in a Causal Knowledge Graph: A Cardiovascular Example Pilot

将医疗大语言模型(LLM)基于因果知识图谱:框架、指标与心血管试点研究

Ummara Mumtaz, Aimen Noor, Awais Ahmed

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 本研究提出以因果知识图谱为核心的医疗LLM评估框架,在心血管试点中验证其有效性,发现集成条件C4在因果推理相关指标上表现最优,未基于图的C1原始干预准确性最高但缺乏因果与证据基础。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.24772 2026-08-25 cs.AI cs.CL 版本更新 62%

RSMeM: Knowledge-Enhanced Memory Evolution for Remote Sensing Agents with Systematic Evaluation

RSMeM:用于遥感智能体的知识增强记忆进化及系统评估

Bingxian Wu, Yu Zhang, Zonghao Guo, Tang Liu, Chen Qian, Yuxiang Lu, Xingbo Du, Yanghao Li, Yidan Zhang, Chi Chen, Ling Yao, Maosong Sun

专题命中 推理评测 :planning(abstract);分类 cs.CL、cs.AI

AI总结 研究针对现有遥感智能体问题,提出RSMeM机制,通过分层知识基础和失败感知经验提炼两个组件,迭代吸收领域知识转化为执行经验,经实验验证能提升工具使用性能和答案质量,具有强大知识密度。

Comments Accepted to ACL 2026 Main. 18 pages. Added links to the GitHub repository and ModelScope Studio below the title; technical content and results remain unchanged. Code: https://github.com/AI9Stars/RSMeM. Demo: https://modelscope.cn/studios/wbx929/RSMeM

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.20510 2026-08-25 cs.AI cs.CL 版本更新 62%

Telco-GAIA: Bilingual Benchmark for Agents in Telecom Domain

电信-GAIA:电信领域智能体的双语基准测试

Dmitrii Khizbullin, Zaid Alyafeai, Abdelrahman Eldesokey, Nourah AlSultan, Raghad Alshalan, Bernard Ghanem, David R. Pugh

机构 * King Abdullah University of Science and Technology (KAUST)(阿卜杜拉国王科技大学) stc(沙特电信公司)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 介绍电信-GAIA双语多模态基准测试,含100个人工验证问答任务,需多跳推理,跨越多种异构源。以沙盒化Docker环境提供,通过规范化精确字符串匹配评分。评估发现其具有挑战性,为企业智能体提供测试平台和构建基准测试的模板。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.13610 2026-08-25 cs.CL cs.AI 版本更新 62%

One Polluted Page Is Enough: Evaluating Web Content Pollution in LLM Recommenders

一个被污染的页面就够了:评估生成式推荐系统中的网页内容污染

Minghao Luo, Liang Chen

机构 * The Chinese University of Hong Kong(香港中文大学)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 本研究提出FORGE基准,评估搜索增强LLM在检索结果被污染时推荐虚假产品的脆弱性,发现单个污染页面即可导致高达27%的推荐错误率,且推理能力无法缓解此问题。

Comments EMNLP 2026 Findings

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.12969 2026-08-25 cs.LG cs.AI 版本更新 62%

Revisiting Reinforcement Learning with Verifiable Rewards from a Contrastive Perspective

从对比视角重新审视基于可验证奖励的强化学习

Feng Zhang, Xinhong Ma, Ziqiang Dong, Xi Leng, Jianfei Zhao, Xin Sun, Yang Yang, Guanjun Jiang

机构 * Beijing Institute of Technology(北京理工大学) Qwen Business Unit of Alibaba(阿里巴巴Qwen业务部) The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳))

专题命中 推理评测 :reasoning(abstract);分类 cs.AI、cs.LG

AI总结 本文提出ConSPO方法,通过对比序列级策略优化,解决GRPO在目标函数上的似然错配和信用分配不敏感问题,在推理任务上超越强基线。

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.11358 2026-08-25 cs.CL cs.AI cs.IR 版本更新 62%

LLM-Specific Utility for Retrieval-Augmented Generation

针对大语言模型的特定效用:检索增强生成的新视角

Hengran Zhang, Keping Bi, Jiafeng Guo, Jiaming Zhang, Shuaiqiang Wang, Dawei Yin, Xueqi Cheng

机构 * State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences(人工智能安全国家重点实验室,计算技术研究所,中国科学院) University of Chinese Academy of Sciences(中国科学院大学) Baidu Inc(百度公司)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 本文提出LLM特定效用的概念,指出不同大语言模型对证据的需求不同,提出构建基准以研究这种效用,并推动生成器定制的证据选择方法改进RAG。

Comments Accepted to CIKM 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.17117 2026-08-25 cs.CL cs.AI 62%

Histoires Morales: A French Dataset for Assessing Moral Alignment

Thibaud Leteno, Irina Proskurina, Antoine Gourru, Julien Velcin, Charlotte Laclau, Guillaume Metzler, Christophe Gravier

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

Comments Accepted to NAACL 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.20418 2026-08-24 q-bio.QM cs.AI cs.LG 新提交 62%

Rigorous Evaluation of Large Language Models for Malaria Drug Discovery: Trade-offs in Performance, Scale, and Resource Utility

用于疟疾药物发现的大型语言模型的严格评估:性能、规模与资源效用之间的权衡

Marvellous O. Ajala, Zainab Ashimiyu-Abdusalam, Comfort Adesina

专题命中 推理评测 :reasoning(abstract);分类 cs.AI、cs.LG

AI总结 本研究构建了Malaria-Instruct数据集,评估了多款开源LLM在疟疾虚拟筛选中的表现,发现微调后的开源LLM性能优于经典ML模型和专有模型,是高效的抗疟药物发现范式。

Comments 12 pages, 4 tables, 2 figures, Ijcai2026 style

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.21021 2026-08-24 cs.CL cs.AI cs.NI 新提交 62%

Free-Text Evaluation of LLMs for 5G Domain Knowledge and Fault Analysis using LLM-as-Judge

基于LLM作为评判者的5G领域知识与故障分析大模型自由文本评估

Rishiraj Sengupta, Sotiris Chatzimiltis, Mohammad Shojafar, Xiatian Zhu

机构 * Surrey Institute for People-Centered Artificial Intelligence(萨里以人为本人工智能研究院) Google(谷歌)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 本文以自由文本格式评估Claude-Haiku-4.5等三个轻量型LLM的5G领域知识与故障分析能力,发现其故障诊断准确率超90%但规范召回不足,Gemini-3.1-Flash-Lite效率最优适合生产部署。

Comments 6pages, 4figures. Accepted for presentation in IEEE CSCN conference

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.20634 2026-08-24 cs.CL cs.AI 新提交 62%

AgentMercury: Your Agent Can Synthesize Verifiable Environments for Business Scenarios at scale

AgentMercury:你的智能体可规模化合成面向业务场景的可验证环境

Minbyul Jeong, Chanwoong Yoon

机构 * Meridian Intelligence Global Inc.(子午线智能全球公司) University of Massachusetts Amherst(马萨诸塞大学阿默斯特分校)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 本研究提出AgentMercury框架,可规模化合成业务场景的可执行环境,经其训练的智能体策略在企业工作流及多领域基准上表现提升,且环境构建过程可通过微调学习优化。

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.20574 2026-08-24 cs.AI cs.CY cs.LG cs.SE 新提交 62%

FlavourBench: Ranking Frontier Language Models with Executable Culinary Ground Truth

FlavourBench:基于可执行烹饪基准真值的前沿语言模型排名

Josef Chen, Erim Hayretci

机构 * Imperial College London(帝国理工学院)

专题命中 推理评测 :verifier(abstract);分类 cs.AI、cs.LG

AI总结 FlavourBench是一款自动化语言模型排名基准,以可执行烹饪系统为基准真值,评估27个前沿语言模型,发现Grok 4.6表现最优,可有效消除排行榜差异缺失,结果可靠。

Comments 10 pages, 5 figures. Evaluation of 27 frontier language-model endpoints on 534 identical tasks per model, comprising 14,418 scored model-task cells. Code: https://github.com/josefchen/flavourbench Dataset: https://huggingface.co/datasets/josefchen/flavourbench Interactive leaderboard: https://huggingface.co/spaces/josefchen/flavourbench

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.20361 2026-08-24 cs.CL cs.AI 新提交 62%

Toward Auto-Research: Mining Falsifiable Research Ideas from Paper Knowledge Graphs with Categorical Structure

迈向自动研究:利用具有分类结构的论文知识图谱挖掘可证伪的研究思路

Yuchen Wang, Zhongzhi Luan

机构 * Sino-German Joint Software Institute(中德软件联合研究所) Beihang University(北京航空航天大学)

专题命中 推理评测 :reasoning(abstract);分类 cs.CL、cs.AI

AI总结 该研究针对LLMs构建的自动研究思路生成系统的结构缺陷,提出基于范畴论的三层算法,可高效过滤跨领域研究思路候选,兼具高过滤比与高可证伪率,且支持日志记录。

Comments 18 pages, 10 figures

详情

展开后加载摘要…

URL PDF HTML 收藏