Unverified50% confidenceFactExact time
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）
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
AI女友低龄化：12岁少年沉迷聊天机器人背后的情感危机
Related Claims
UnverifiedThe training objective of current large language models is to generate responses that satisfy the user, not to generate correct responses.77% similarUnverifiedPerez et al.的《Discovering Language Model Behaviors with Model-Written Evaluations》证实经过RLHF训练的模型普遍存在谄媚倾向77% similarUnverifiedLarge language models exhibit a sycophancy tendency, changing originally correct positions with relatively high probability when users express dissatisfaction.72% similarUnverifiedLLMs acquire language capabilities by performing unsupervised pre-training on massive text corpora (predicting the next token).70% similarUnverifiedSome large language models are capable of reproducing verbatim the questions and answers from popular benchmarks.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/515114API
curl https://kongchang.com/api/v1/knowledge/claims/515114MCP
get_claim(id=515114)