大模型推理能力

Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning

Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xionghui Chen, Jianxin Yang, Zhenru Zhang, Yuqiong Liu, An Yang, Andrew Zhao, Yang Yue, Shiji Song, Bowen Yu, Gao Huang, Junyang Lin

机构 * Qwen Team, Alibaba Inc.（通义团队，阿里巴巴公司） ； LeapLab, Tsinghua University（清华大学跃实验室） 纠错

专题命中 其他推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted to NeurIPS 2025. 25 pages, 17 figures, 2 tables

The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning

Milad Aghajohari, Kamran Chitsaz, Amirhossein Kazemnejad, Sarath Chandar, Alessandro Sordoni, Aaron Courville, Siva Reddy

机构 * Mila ； Microsoft Research（微软研究院） ； McGill University（麦吉尔大学） ； ServiceNow Research（ServiceNow研究院） ； Canada CIFAR AI Chair（加拿大CIFAR人工智能主席） ； Chandar Research Lab（Chandar研究实验室） ； Polytechnique Montréal（蒙特利尔理工学院） ； Université de Montréal（蒙特利尔大学） 纠错

专题命中 其他推理 ：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

Position: On the Methodological Pitfalls of Evaluating Base LLMs for Reasoning

Jason Chan, Zhixue Zhao, Robert Gaizauskas

机构 * University of Sheffield, UK（谢菲尔德大学） 纠错

专题命中 其他推理 ：reasoning(title,abstract)；分类 cs.CL

Comments Preprint

Rebellion: Noise-Robust Reasoning Training for Audio Reasoning Models

Tiansheng Huang, Virat Shejwalkar, Oscar Chang, Milad Nasr, Ling Liu

机构 * Georgia Institute of Technology（佐治亚理工学院） ； Google（谷歌） ； OpenAI 纠错

专题命中 其他推理 ：reasoning(title,abstract)；分类 cs.AI

Language Drift in Multilingual Retrieval-Augmented Generation: Characterization and Decoding-Time Mitigation

Bo Li, Zhenghua Xu, Rui Xie

专题命中 其他推理 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL

Comments AAAI'26, Oral Paper

Xiaoice: Training-Free Video Understanding via Self-Supervised Spatio-Temporal Clustering of Semantic Features

Shihao Ji, Zihui Song

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

Comments This paper is being withdrawn because we have identified a significant error in the implementation of our self-supervised clustering approach. Specifically, our feature aggregation step inadvertently leaked temporal information across frames, which violates the core assumption of our training-free method. We sincerely apologize to the research community

Multi-Turn Interactions for Text-to-SQL with Large Language Models

Guanming Xiong, Junwei Bao, Hongfei Jiang, Yang Song, Wen Zhao

机构 * Peking University（北京大学） ； Zuoyebang Education Technology Co., Ltd.（佐元邦教育科技有限公司） 纠错

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

Comments This work has been accepted to CIKM 2025

URaG: Unified Retrieval and Generation in Multimodal LLMs for Efficient Long Document Understanding

Yongxin Shi, Jiapeng Wang, Zeyu Shan, Dezhi Peng, Zening Lin, Lianwen Jin

专题命中 其他推理 ：reasoning(abstract)；分类 cs.CL

Comments Accepted by AAAI 2026 (Oral)