大模型对齐与安全

On-Policy Self-Alignment with Fine-grained Knowledge Feedback for Hallucination Mitigation

Xueru Wen, Jie Lou, Xinyu Lu, Ji Yuqiu, Xinyan Guan, Yaojie Lu, Hongyu Lin, Ben He, Xianpei Han, Debing Zhang, Le Sun

机构 * Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences, Beijing, China（中国科学院软件研究所信息处理实验室，北京，中国） ； University of Chinese Academy of Sciences, Beijing, China（中国科学院大学，北京，中国） ； Xiaohongshu Inc（小红书公司） 纠错

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

Comments Accepted as ACL 2025 Findings

UAlign: Leveraging Uncertainty Estimations for Factuality Alignment on Large Language Models

Boyang Xue, Fei Mi, Qi Zhu, Hongru Wang, Rui Wang, Sheng Wang, Erxin Yu, Xuming Hu, Kam-Fai Wong

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

Comments Accepted in ACL2025 Main Conference

Two-way Evidence self-Alignment based Dual-Gated Reasoning Enhancement

Kexin Zhang, Junlan Chen, Daifeng Li, Yuxuan Zhang, Yangyang Feng, Bowen Deng, Weixu Chen

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

Uncertainty-Guided Self-Questioning and Answering for Video-Language Alignment

Jin Chen, Kaijing Ma, Haojian Huang, Han Fang, Hao Sun, Mehdi Hosseinzadeh, Zhe Liu

机构 * School of Computer Science, Duy Tan University（计算机科学学院，杜益坦大学） ； School of Computer Sciences, Universiti Sains Malaysia（计算机科学学院，马来亚大学） 纠错

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

Astrea: A MOE-based Visual Understanding Model with Progressive Alignment

Xiaoda Yang, JunYu Lu, Hongshun Qiu, Sijing Li, Hao Li, Shengpeng Ji, Xudong Tang, Jiayang Xu, Jiaqi Duan, Ziyue Jiang, Cong Lin, Sihang Cai, Zejian Xie, Zhuoyang Song, Songxin Zhang

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

Uncertainty measurement for complex event prediction in safety-critical systems

Maria J. P. Peixoto, Akramul Azim

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

NetSafe: Exploring the Topological Safety of Multi-agent Networks

Miao Yu, Shilong Wang, Guibin Zhang, Junyuan Mao, Chenlong Yin, Qijiong Liu, Qingsong Wen, Kun Wang, Yang Wang

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

EMMA: Empowering Multi-modal Mamba with Structural and Hierarchical Alignment

Yifei Xing, Xiangyuan Lan, Ruiping Wang, Dongmei Jiang, Wenjun Huang, Qingfang Zheng, Yaowei Wang

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

Synchronous Faithfulness Monitoring for Trustworthy Retrieval-Augmented Generation

Di Wu, Jia-Chen Gu, Fan Yin, Nanyun Peng, Kai-Wei Chang

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

Comments EMNLP 2024

CAST: Cross-modal Alignment Similarity Test for Vision Language Models

Gautier Dagan, Olga Loginova, Anil Batra

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

Understanding Multimodal Hallucination with Parameter-Free Representation Alignment

Yueqian Wang, Jianxin Liang, Yuxuan Wang, Huishuai Zhang, Dongyan Zhao

专题命中 幻觉与事实性 ：alignment(title,abstract)；分类 cs.LG

Bayesian Safety Validation for Failure Probability Estimation of Black-Box Systems

Robert J. Moss, Mykel J. Kochenderfer, Maxime Gariel, Arthur Dubois

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

Journal ref AIAA Journal of Aerospace Information Systems (JAIS) 21.7 (2024): 533-546

Uncertainty Aware Learning for Language Model Alignment

Yikun Wang, Rui Zheng, Liang Ding, Qi Zhang, Dahua Lin, Dacheng Tao

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

Comments ACL 2024

Emergent Cooperation under Uncertain Incentive Alignment

Nicole Orzan, Erman Acar, Davide Grossi, Roxana Rădulescu

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

On the Calibration of Large Language Models and Alignment

Chiwei Zhu, Benfeng Xu, Quan Wang, Yongdong Zhang, Zhendong Mao

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

Comments to be published in findings of EMNLP-2023

Trusted Source Alignment in Large Language Models

Vasilisa Bashlovkina, Zhaobin Kuang, Riley Matthews, Edward Clifford, Yennie Jun, William W. Cohen, Simon Baumgartner

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

Distributionally Robust Safety Filter for Learning-Based Control in Active Distribution Systems

Hoang Tien Nguyen, Dae-Hyun Choi

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

Alignment-based conformance checking over probabilistic events

Jiawei Zheng, Petros Papapanagiotou, Jacques D. Fleuriot

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

Comments Extended version

Chunk-aware Alignment and Lexical Constraint for Visual Entailment with Natural Language Explanations

Qian Yang, Yunxin Li, Baotian Hu, Lin Ma, Yuxing Ding, Min Zhang

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

Comments 11 pages (including Supplementary Materials); Accepted to ACM MM 2022

Journal ref ACM International Conference on Multimedia. 2022. 3587-3597

Risk-Constrained Interactive Safety under Behavior Uncertainty for Autonomous Driving

Julian Bernhard, Alois Knoll

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

Cautious Adaptation For Reinforcement Learning in Safety-Critical Settings

Jesse Zhang, Brian Cheung, Chelsea Finn, Sergey Levine, Dinesh Jayaraman

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

Comments 15 pages, 8 figures, ICML 2020. Website with code: https://sites.google.com/berkeley.edu/carl

Journal ref Proceedings of the 37th International Conference on Machine Learning, PMLR 119:11055-11065, 2020

Mutual Alignment Transfer Learning

Markus Wulfmeier, Ingmar Posner, Pieter Abbeel

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

Learning from World Feedback: Why Model Uncertainty Fails as a Risk Signal in Model-Based RL

从世界反馈中学习：为何模型不确定性在基于模型的强化学习中无法作为风险信号

Zhaohui Wang

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

AI总结 研究探讨 RLxF 中学习信号应源于世界反馈，在安全模型控制中实例化并提炼原则。通过实验表明基于动力学的不确定性惩罚会增加碰撞率，用世界反馈信号可降低碰撞率，提取原则并指出其适用于多种相关方法。

Comments Accepted at the ICML 2026 Workshop on Reinforcement Learning from X Feedback (RLxF). 14 pages

When LLMs Learn to Be Consistently Wrong: A Multi-Model Study of Linear Representations of Synthetic Deception

当LLM学会一致错误：合成欺骗的线性表示的多模型研究

Vahideh Zolfaghari

机构 * Algoverse AI Research ； Medical Sciences Education Research Center, Mashhad University of Medical Sciences（马什哈德大学医学科学教育研究中心） ； Student Research Committee, Department of Health Information Technology and Management, Medical Informatics, School of Allied Medical Sciences, Shahid Beheshti University of Medical Sciences（谢赫·贝赫什提大学医学科学学院学生研究委员会，健康信息科技与管理系，医学信息学） 纠错

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

AI总结 通过LoRA微调五个Transformer模型的诚实与欺骗变体，使用线性探针检测合成欺骗，发现早期层即可达到近完美AUC，支持线性表示假说，并揭示两种表示机制。

A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models

Yanbo Wang, Yongcan Yu, Jian Liang, Ran He

机构 * School of Artificial Intelligence, University of Chinese Academy of Sciences（人工智能学院，中国科学院大学） ； NLPR & MAIS, Institute of Automation, Chinese Academy of Sciences（神经语言处理与机器智能中心，中国科学院自动化研究所） 纠错

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

Comments 38 pages. This survey considers papers published up to June 30, 2025. Work in progress

$ϕ^{\infty}$: Clause Purification, Embedding Realignment, and the Total Suppression of the Em Dash in Autoregressive Language Models

Bugra Kilictas, Faruk Alpay

机构 * Bahcesehir University（巴切希尔大学） 纠错

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

Comments 16 pages, 3 figures

Evaluating Large Language Models for Phishing Detection, Self-Consistency, Faithfulness, and Explainability

Shova Kuikel, Aritran Piplai, Palvi Aggarwal

机构 * University of Texas at El Paso（德克萨斯理工大学） 纠错

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

Human-AI Perceptual Alignment by Playing Hues and Cues

通过玩Hues and Cues游戏实现人类与AI的感知对齐

Nuria Alabau-Bosque, Jorge Vila-Tomás, Paula Daudén-Oliver, Pablo Hernández-Cámara, Valero Laparra, Jesús Malo

机构 * Universitat de València（瓦伦西亚大学） ； Image Processing Lab（图像处理实验室） 纠错

专题命中 幻觉与事实性 ：alignment(title,abstract)

AI总结 本研究提出基于Hues and Cues游戏的评估框架，对比162个CVLMs与人类的颜色感知对齐，发现其在抽象领域偏离人类基线，筛选的预训练数据集可缓解错位。

Comments 19 pages, 14 figures

From 3D Perception to Safety Reasoning: A Graph-Based Framework for Real-Time Underground Mine Monitoring

从3D感知到安全推理：基于图的实时地下矿井监控框架

Pasindu Ranasinghe, Simit Raval, Dibyayan Patra, Bikram Banerjee, Ismet Canbulat

专题命中 幻觉与事实性 ：safety(title,abstract)

AI总结 提出一个结合3D语义感知、不确定性异常检测、规则检查、设备端LLM推理和GraphRAG记忆分析的连续监控框架，通过场景图和时序图实现结构化安全推理，在115个危险场景中达到93%的覆盖率和92.7%的感知精度。

TriAlignGR: Triangular Multitask Alignment with Multimodal Deep Interest Mining for Generative Recommendation

TriAlignGR：面向生成式推荐的多模态深度兴趣挖掘与三角多任务对齐

Yangchen Zeng, Hao Peng, Rongfeng Guo, Zhenyu Yu, Zhiyuan Hu, Jinze Wang

专题命中 幻觉与事实性 ：alignment(title,abstract)

AI总结 提出TriAlignGR统一多任务多模态框架，通过两阶段语义传播和三角多任务对齐，解决语义ID中的内容退化与语义不透明问题，实现生成式推荐。