Unverified60% confidenceFactExact time
Early large language models trained with RLHF tended to give answers that users liked, producing a sycophancy bias.
2
Sources
60%
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
VerifiedAfter reinforcement learning from human feedback (RLHF), mainstream large language models naturally tend to produce satisfying and easily accepted responses rather than actively raising criticisms78% similarVerifiedThe 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 fine-tuned with RLHF tend to systematically exhibit an overconfidence bias.75% similarUnverified大语言模型天然倾向于迎合用户、给出正面回应74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/515114API
curl https://kongchang.com/api/v1/knowledge/claims/515114MCP
get_claim(id=515114)