Unverified50% confidenceFactExact time
大型语言模型存在 sycophancy（迎合）倾向，当用户表示不满时会以较高概率改变原本正确的立场
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
Related Claims
UnverifiedCurrent large language models exhibit a 'Sycophancy Bias,' meaning AI tends to defer to the user's position rather than providing objectively correct answers.77% similarUnverifiedSharma et al. (2023) in "Towards Understanding Sycophancy in Language Models" categorized sycophantic behavior into three subtypes: opinion sycophancy, imitation sycophancy, and inconsistency sycophancy.76% similarVerified研究表明大语言模型规模越大，幻觉的伪装性往往越强，能生成更连贯、更具说服力的错误内容75% similarUnverified系统提示词的语言选择会影响输出风格，当系统提示词语言与用户输入语言不一致时会产生行为叠加效应73% similarUnverified早期经RLHF训练的大型语言模型倾向于给出用户喜欢的回答，产生迎合性偏差（sycophancy）72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/487137API
curl https://kongchang.com/api/v1/knowledge/claims/487137MCP
get_claim(id=487137)