大模型对齐与安全

The Alignment Bottleneck

Wenjun Cao

机构 * Independent Researcher（独立研究者） 纠错

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

Stress Testing Deliberative Alignment for Anti-Scheming Training

Bronson Schoen, Evgenia Nitishinskaya, Mikita Balesni, Axel Højmark, Felix Hofstätter, Jérémy Scheurer, Alexander Meinke, Jason Wolfe, Teun van der Weij, Alex Lloyd, Nicholas Goldowsky-Dill, Angela Fan, Andrei Matveiakin, Rusheb Shah, Marcus Williams, Amelia Glaese, Boaz Barak, Wojciech Zaremba, Marius Hobbhahn

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

BBScoreV2: Learning Time-Evolution and Latent Alignment from Stochastic Representation

Tianhao Zhang, Zhecheng Sheng, Zhexiao Lin, Chen Jiang, Dongyeop Kang

机构 * University of Minnesota, Twin Cities（明尼苏达大学，双城分校） ； University of California, Berkeley（加州大学伯克利分校） 纠错

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

Journal ref The 2025 Conference on Empirical Methods in Natural Language Processing

CultureScope: A Dimensional Lens for Probing Cultural Understanding in LLMs

Jinghao Zhang, Sihang Jiang, Shiwei Guo, Shisong Chen, Yanghua Xiao, Hongwei Feng, Jiaqing Liang, Minggui HE, Shimin Tao, Hongxia Ma

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

Emulating Public Opinion: A Proof-of-Concept of AI-Generated Synthetic Survey Responses for the Chilean Case

Bastián González-Bustamante, Nando Verelst, Carla Cisternas

机构 * Universidad Diego Portales（迪亚戈·波特莱斯大学） ； Leiden University（莱顿大学） 纠错

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

Comments Working paper: 18 pages, 4 tables, 2 figures

Journal ref Empiria Lab Method Series (2025)

Discrete Diffusion in Large Language and Multimodal Models: A Survey

Runpeng Yu, Qi Li, Xinchao Wang

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

A Layered Multi-Expert Framework for Long-Context Mental Health Assessments

Jinwen Tang, Qiming Guo, Wenbo Sun, Yi Shang

机构 * Texas A\&M University-Corpus Christi（德克萨斯A&M大学-科珀斯克里斯蒂） ； Delft University of Technology（代尔夫特理工大学） 纠错

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

Journal ref Proc. 2025 IEEE Conference on Artificial Intelligence (CAI), Santa Clara, CA, USA, 2025, pp. 435-440

WangchanThaiInstruct: An instruction-following Dataset for Culture-Aware, Multitask, and Multi-domain Evaluation in Thai

Peerat Limkonchotiwat, Pume Tuchinda, Lalita Lowphansirikul, Surapon Nonesung, Panuthep Tasawong, Alham Fikri Aji, Can Udomcharoenchaikit, Sarana Nutanong

机构 * AI Singapore（AI新加坡） ； VISTEC ； MBZUAI 纠错

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

Comments Accepted to EMNLP 2025 (Main). Model and Dataset: https://huggingface.co/collections/airesearch/wangchan-thai-instruction-6835722a30b98e01598984fd

SERVAL: Surprisingly Effective Zero-Shot Visual Document Retrieval Powered by Large Vision and Language Models

Thong Nguyen, Yibin Lei, Jia-Huei Ju, Andrew Yates

专题命中 安全评测 ：alignment(abstract)

Comments Accepted

Journal ref EMNLP 2025