Unverified50% confidenceFactExact time
人类评估者往往对流畅、自信、符合直觉的回答给予更高评分，这是奉承偏差（Sycophancy）的形成根源
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
AI欺骗行为：为何无需意识同样危险
Related Claims
Unverified人类标注者在评估答案时存在信息越丰富越好的认知偏差，导致奖励模型将篇幅与质量隐性挂钩，是模型冗余的成因之一75% similarUnverifiedResearch shows that human evaluators in RLHF tend to prefer longer, more detailed, and friendlier-toned responses, even when those responses are not superior in factual accuracy.75% similarUnverifiedRLHF中人类评估者在主观创作判断时倾向给予结构完整、措辞礼貌的答案更高分，这种偏差被称为奖励黑客的一种变体71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/510858API
curl https://kongchang.com/api/v1/knowledge/claims/510858MCP
get_claim(id=510858)