Mechanics of Bias and Reasoning: Interpreting the Impact of Chain-of-Thought Prompting on Gender Bias in LLMs
偏见与推理的力学:分析链式推理提示对大语言模型中性别偏见的影响
机构 * Mila – Quebec AI Institute(魁北克人工智能研究所) ; McGill University(麦吉尔大学) ; HEC Montreal(蒙特利尔HEC)
专题命中 推理评测 :reasoning(title,abstract);chain-of-thought(title,abstract);CoT(abstract,abstract_cn);分类 cs.CL、cs.AI
AI总结 本文研究了链式推理提示对大语言模型中性别偏见的影响,结合基准测试评估与机制可解释性技术,发现链式推理并未有效减少偏见,偏见仍存在于隐藏表示中。
Comments 24 pages, 6 figures, including appendix. Accepted at the ICLR 2026 Workshop on Algorithmic Fairness Across Alignment Procedures and Agentic Systems. Submitted to COLM 2026