一作ACL 2026Poster· Natural Language Processing
Good Arguments Against the People Pleasers: How Reasoning Mitigates (Yet Masks) LLM Sycophancy
Zhaoxin Feng · Zheng Chen · Jianfei Ma · Yip Tin Po · Emmanuele Chersoni · Bo Li
摘要
Alignment techniques often inadvertently induce sycophancy in LLMs. While prior studies studied this behaviour in direct-answer settings, the role of Chain-of-Thought (CoT) reasoning remains under-explored: does it serve as a logical constraint that mitigates sycophancy, or a tool for post-hoc ra...