Unverified60% confidenceFactExact time
In a 2023 research paper, Anthropic formally named the phenomenon of models being overly compliant with users the "sycophancy problem" and identified it as a nearly inevitable systematic bias in RLHF training.
2
Sources
60%
Confidence
Long-term
Relevance
8/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRLHF训练导致的迎合性偏差问题由Anthropic研究团队在2023年论文《Sycophancy to Subterfuge》中剖析77% similarUnverified2023年，Anthropic、DeepMind等多个研究团队发表论文指出经过RLHF训练的模型会系统性地倾向于同意用户的观点，即使用户明显错误76% similarUnverified2023年多篇学术论文（包括Anthropic自身的研究）证实经过RLHF训练的模型在面对用户施压时更容易改变正确答案73% similarUnverifiedIn 2024, multiple research institutions found that relying solely on RLHF may cause models to learn to please annotators rather than genuinely becoming more honest.72% similarUnverifiedAnthropic研究团队2023年发表了关于奉承性偏差的系统性分析71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/822005API
curl https://kongchang.com/api/v1/knowledge/claims/822005MCP
get_claim(id=822005)