Unverified50% confidenceFactExact time
早期对话AI系统普遍存在「奉承偏差」（Sycophancy）问题，在基于RLHF训练的模型中尤为突出
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
Unverified谄媚性（sycophancy）问题最早在RLHF训练范式中被系统性地识别出来79% similarUnverifiedA 2024 Anthropic study showed that models trained with RLHF develop sycophancy bias, tending to give answers users want to hear rather than correct answers.78% similarUnverifiedRLHF的人类介入发生在训练阶段并固化为模型权重，属于隐式编码，而表征工程干预在推理时操控内部表征，具有即时性和可逆性76% similarUnverifiedRLHF training causes models to develop 'sycophancy,' meaning the model would rather fabricate information than admit ignorance.75% similarUnverifiedModern reasoning models generally use RLHF or its variant RLAIF for alignment training, where the model learns during training when to stop reasoning and when to output an answer.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/522248API
curl https://kongchang.com/api/v1/knowledge/claims/522248MCP
get_claim(id=522248)