大模型对齐与安全

VaPR -- Vision-language Preference alignment for Reasoning

Rohan Wadhawan, Fabrice Y Harel-Canada, Zi-Yi Dou, Suhaila Shakiah, Robinson Piramuthu, Nanyun Peng

机构 * Department of Computer Science, University of California Los Angeles（加州大学洛杉矶分校计算机科学系） ； Amazon.com, Inc.（亚马逊公司） 纠错

专题命中 偏好对齐 ：alignment(title)；DPO(abstract)；分类 cs.AI、cs.LG

Journal ref COLM 2025

Efficient Training of Robust Traditional Chinese LLaMA-1B on a Single Consumer GPU: Continual Pre-training, SFT, and DPO

Yu-Cheng Chih, Ming-Tao Duan, Yong-Hao Hou

专题命中 偏好对齐 ：DPO(title,abstract)；分类 cs.CL

Comments 17 pages, 1 figures, 2 tables. Technical report. Introduces PureTC-1B, an adapter-based pipeline for stabilizing Small Language Models in Traditional Chinese using CPT, SFT, and DPO

Differential Information Distribution: A Bayesian Perspective on Direct Preference Optimization

Yunjae Won, Hyunji Lee, Hyeonbin Hwang, Minjoon Seo

机构 * KAIST AI（韩国科学技术院人工智能研究所） 纠错

专题命中 偏好对齐 ：DPO(abstract,comments)；alignment(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Preprint, under review. 39 pages, 12 figures. Updates from v1: Added new theoretical results on DPO training dynamics and policy exploration, included experiments with Qwen3-4B, and refined the discussion of log-margin dynamics

Enhancing Personalized Multi-Turn Dialogue with Curiosity Reward

Yanming Wan, Jiaxing Wu, Marwa Abdulhai, Lior Shani, Natasha Jaques

机构 * Google DeepMind（谷歌DeepMind） ； University of Washington（华盛顿大学） ； Google Research（谷歌研究） ； University of California, Berkeley（伯克利加州大学） 纠错

专题命中 偏好对齐 ：RLHF(abstract)；safety(abstract)；分类 cs.CL、cs.AI

Out-of-Distribution Detection using Synthetic Data Generation

Momin Abbas, Muneeza Azmat, Raya Horesh, Mikhail Yurochkin

专题命中 偏好对齐 ：RLHF(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted to COLM 2025. Camera-ready version

Optimal Stopping vs Best-of-$N$ for Inference Time Optimization

Yusuf Kalayci, Vinod Raman, Shaddin Dughmi

机构 * University of Southern California（南加州大学） ； University of Michigan（密歇根大学） 纠错

专题命中 偏好对齐 ：RLHF(abstract)；分类 cs.CL、cs.LG

Comments 24 pages

GRPO++: Enhancing Dermatological Reasoning under Low Resource Settings

Ismam Nur Swapnil, Aranya Saha, Tanvir Ahmed Khan, Mohammad Ariful Haque

机构 * Department of Electrical and Electronic Engineering, Bangladesh University of Engineering and Technology (BUET)（电气与电子工程系，孟加拉国工程与技术大学） 纠错

专题命中 偏好对齐 ：DPO(abstract)；分类 cs.CL、cs.LG

Comments Will be submitted at IEEE JBHI

FANS -- Formal Answer Selection for Natural Language Math Reasoning Using Lean4

Jiarui Yao, Ruida Wang, Tong Zhang

机构 * University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校） 纠错

专题命中 偏好对齐 ：trustworthy(abstract)；分类 cs.CL、cs.AI

StepORLM: A Self-Evolving Framework With Generative Process Supervision For Operations Research Language Models

Chenyu Zhou, Tianyi Xu, Jianghao Lin, Dongdong Ge

机构 * Shanghai Jiao Tong University（上海交通大学） 纠错

专题命中 偏好对齐 ：DPO(abstract)；分类 cs.AI

Diffusion Model as a Noise-Aware Latent Reward Model for Step-Level Preference Optimization

Tao Zhang, Cheng Da, Kun Ding, Huan Yang, Kun Jin, Yan Li, Tingting Gao, Di Zhang, Shiming Xiang, Chunhong Pan

机构 * MAIS, CASIA（中国科学院自动化研究所MAIS实验室） ； Kuaishou Technology（快手科技） ； School of Artificial Intelligence, UCAS（中国科学院大学人工智能学院） 纠错

专题命中 偏好对齐 ：alignment(abstract)

Comments NeurIPS 2025

UpSafe$^\circ$C: Upcycling for Controllable Safety in Large Language Models

Yuhao Sun, Zhuoer Xu, Shiwen Cui, Kun Yang, Lingyun Yu, Yongdong Zhang, Hongtao Xie

专题命中 安全训练 ：safety(title,abstract)；alignment(abstract)；jailbreak(abstract)；分类 cs.AI、cs.LG

Predictive Modeling and Explainable AI for Veterinary Safety Profiles, Residue Assessment, and Health Outcomes Using Real-World Data and Physicochemical Properties

Hossein Sholehrasa, Xuan Xu, Doina Caragea, Jim E. Riviere, Majid Jaberi-Douraki

机构 * DATA Consortium and FARAD Program, Kansas State University, Olathe, KS, USA（1DATA Consortium和FARAD计划，堪萨斯州立大学） ； Department of Computer Science, Kansas State University, Manhattan, KS, USA（计算机科学系，堪萨斯州立大学） ； Department of Statistics, Kansas State University, Manhattan, KS, USA（统计学系，堪萨斯州立大学） ； Department of Mathematics, Kansas State University, Olathe, KS, USA（数学系，堪萨斯州立大学） 纠错

专题命中 安全训练 ：safety(title,abstract)；分类 cs.AI、cs.LG

In AI Sweet Harmony: Sociopragmatic Guardrail Bypasses and Evaluation-Awareness in OpenAI gpt-oss-20b

Nils Durner

专题命中 安全训练 ：safety(abstract)；分类 cs.CL、cs.AI

Comments 27 pages, 1 figure

CRUST-Bench: A Comprehensive Benchmark for C-to-safe-Rust Transpilation

Anirudh Khatry, Robert Zhang, Jia Pan, Ziteng Wang, Qiaochu Chen, Greg Durrett, Isil Dillig

机构 * The University of Texas at Austin（德克萨斯大学奥斯汀分校） ； New York University（纽约大学） 纠错

专题命中 安全训练 ：safety(abstract)；分类 cs.CL、cs.LG

Comments To be published at COLM, 2025

LLM-Based Multi-Task Bangla Hate Speech Detection: Type, Severity, and Target

Md Arid Hasan, Firoj Alam, Md Fahad Hossain, Usman Naseem, Syed Ishtiaque Ahmed

机构 * University of Toronto（多伦多大学） ； Qatar Computing Research Institute（卡塔尔计算研究院） ； Daffodil International University（达芙妮国际大学） ； Macquarie University（麦考瑞大学） 纠错

专题命中 安全训练 ：safety(abstract)；分类 cs.CL

VLA-R1: Enhancing Reasoning in Vision-Language-Action Models

Angen Ye, Zeyu Zhang, Boyuan Wang, Xiaofeng Wang, Dapeng Zhang, Zheng Zhu

机构 * GigaAI ； CASIA ； Tsinghua University（清华大学） 纠错

专题命中 安全训练 ：alignment(abstract)

AdvEvo-MARL: Shaping Internalized Safety through Adversarial Co-Evolution in Multi-Agent Reinforcement Learning

Zhenyu Pan, Yiting Zhang, Zhuo Liu, Yolo Yunlong Tang, Zeliang Zhang, Haozheng Luo, Yuwei Han, Jianshu Zhang, Dennis Wu, Hong-Yu Chen, Haoran Lu, Haoyang Fang, Manling Li, Chenliang Xu, Philip S. Yu, Han Liu

机构 * Northwestern University（西北大学） ； University of Illinois at Chicago（伊利诺伊大学香槟分校） ； University of Rochester（罗切斯特大学） ； Carnegie Mellon University（卡内基梅隆大学） 纠错

专题命中 越狱攻击 ：safety(title,abstract)；jailbreak(abstract)；分类 cs.AI

Inverse Language Modeling towards Robust and Grounded LLMs

Davide Gabrielli, Simone Sestito, Iacopo Masi

机构 * Sapienza University of Rome（罗马萨皮恩扎大学） ； OmnAI Lab（OmnAI实验室） 纠错

专题命中 红队测试 ：red teaming(abstract)；trustworthy(abstract)；分类 cs.CL

Defend LLMs Through Self-Consciousness

Boshi Huang, Fabio Nonato de Paula

专题命中 提示注入 ：prompt injection(abstract)；分类 cs.CL、cs.AI

Comments company requests to withdraw

OpenAI's GPT-OSS-20B Model and Safety Alignment Issues in a Low-Resource Language

Isa Inuwa-Dutse

机构 * University of Huddersfield（赫德菲尔德大学） 纠错

专题命中 幻觉与事实性 ：alignment(title,abstract)；safety(title,abstract)；分类 cs.CL、cs.AI

Comments 6 pages, 4 tables

Software Engineering for Self-Adaptive Robotics: A Research Agenda

Hassan Sartaj, Shaukat Ali, Ana Cavalcanti, Lukas Esterle, Cláudio Gomes, Peter Gorm Larsen, Anastasios Tefas, Jim Woodcock, Houxiang Zhang

机构 * Simula Research Laboratory（Simula研究实验室） ； University of York（约克大学） ； Aarhus University（奥胡斯大学） ； Aristotle University of Thessaloniki（塞萨洛尼基亚里士多德大学） ； Southwest University（西南大学） ； Norwegian University of Science and Technology（挪威科学与技术大学） 纠错

专题命中 幻觉与事实性 ：safety(abstract)；trustworthy(abstract)

Microsaccade-Inspired Probing: Positional Encoding Perturbations Reveal LLM Misbehaviours

Rui Melo, Rui Abreu, Corina S. Pasareanu

机构 * Carnegie Mellon University（卡内基梅隆大学） ； FEUP（费拉尔大学） ； INESC-ID（葡萄牙里斯本信息技术与创新研究中心） 纠错

专题命中 幻觉与事实性 ：safety(abstract)；分类 cs.AI、cs.LG

Comments 9 main pages, 13 appendix pages

Confidence-Aware Routing for Large Language Model Reliability Enhancement: A Multi-Signal Approach to Pre-Generation Hallucination Mitigation

Nandakishor M

机构 * AI Safety Research（人工智能安全研究） ； Convai Innovations（Convai创新） 纠错

专题命中 幻觉与事实性 ：alignment(abstract)；分类 cs.CL、cs.AI

Design and Application of Multimodal Large Language Model Based System for End to End Automation of Accident Dataset Generation

MD Thamed Bin Zaman Chowdhury, Moazzem Hossain

专题命中 幻觉与事实性 ：safety(abstract)；分类 cs.CL

Comments This paper is accepted for presentation in TRB annual meeting 2026. The version presented here is the preprint version before peer review process

TrimTokenator: Towards Adaptive Visual Token Pruning for Large Multimodal Models

Hao Zhang, Mengsi Lyu, Chenrui He, Yulong Ao, Yonghua Lin

机构 * Beijing Academy of Artificial Intelligence (BAAI)（北京人工智能研究院） 纠错

专题命中 幻觉与事实性 ：alignment(abstract)

Comments 15 pages

ZK-WAGON: Imperceptible Watermark for Image Generation Models using ZK-SNARKs

Aadarsh Anantha Ramakrishnan, Shubham Agarwal, Selvanayagam S, Kunwar Singh

专题命中 隐私与版权 ：trustworthy(abstract)；分类 cs.AI

Comments Accepted at AI-ML Systems 2025, Bangalore, India, https://www.aimlsystems.org/2025/

TriAlignXA: An Explainable Trilemma Alignment Framework for Trustworthy Agri-product Grading

Jianfei Xie, Ziyang Li

机构 * School of Software, Xinjiang University（软件学院，新疆大学） 纠错

专题命中 安全评测 ：trustworthy(title,abstract)；alignment(title)；分类 cs.CY

Bridging Collaborative Filtering and Large Language Models with Dynamic Alignment, Multimodal Fusion and Evidence-grounded Explanations

Bo Ma, LuYao Liu, Simon Lau, Chandler Yuan, and XueY Cui, Rosie Zhang

机构 * Department of Software \& Microelectronics, Peking University, Beijing, China ； Economic Law School, China University of Political Science ； Financial Media, Peking University, ChangSha, China 纠错

专题命中 安全评测 ：alignment(title)；trustworthy(abstract)；分类 cs.CL、cs.AI

Trustworthy Summarization via Uncertainty Quantification and Risk Awareness in Large Language Models

Shuaidong Pan, Di Wu

专题命中 安全评测 ：trustworthy(title,abstract)；分类 cs.CL、cs.AI

A Flexible Method for Behaviorally Measuring Alignment Between Human and Artificial Intelligence Using Representational Similarity Analysis

Mattson Ogg, Ritwik Bose, Jamie Scharf, Christopher Ratto, Michael Wolmetz

专题命中 安全评测 ：alignment(title,abstract)；分类 cs.AI