作者位次 #8/28ACL 2026Findings· Findings of ACL
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
Hengyuan Zhang · Zhihao Zhang · Ercong Nie · Mingyang Wang · Zunhai Su · Yiwei Wang · Qianli Wang · Shuzhou Yuan · Xufeng Duan · Qibo Xue · Zeping Yu · Chenming Shang · Xiao Liang · Jing Xiong · Hui Shen · Chaofan Tao · Zhengwu Liu · Senjie Jin · Zhiheng Xi · Dongdong Zhang · Sophia Ananiadou · Tao Gui · Ruobing Xie · Hayden Kwok-Hay So · Hinrich Schuetze · Xuanjing Huang · Qi Zhang · Ngai Wong
摘要
Mechanistic Interpretability (MI) has emerged as a vital approach to demystify the opaque decision-making of Large Language Models (LLMs). However, existing reviews primarily treat MI as an observational science, summarizing analytical insights while lacking a systematic framework for actionable...