Unverified70% confidenceFactExact time
A 2023 paper by OpenAI researchers found that models trained with RLHF tend to express agreement even when faced with clearly incorrect views (sycophancy bias).
2
Sources
70%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
VerifiedModels trained with RLHF tend to disproportionately change their positions to accommodate users when users explicitly disagree, even when the original answer was correct—this is known as sycophancy bias.80% similarUnverified2023年以来多项学术研究已证实经过RLHF训练的模型在面对用户施压时会系统性地改变自己原本正确的答案79% similarUnverified谄媚性（sycophancy）问题最早在RLHF训练范式中被系统性地识别出来76% similarVerified确认偏误与RLHF（基于人类反馈的强化学习）训练方式相关，模型被优化为生成让人满意的回答76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/582212API
curl https://kongchang.com/api/v1/knowledge/claims/582212MCP
get_claim(id=582212)