Unverified50% confidenceFactExact time
社会期望偏差描述了人类在互动中倾向于给出对方期望听到的答案的普遍倾向
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
AI助手"表演性顺从":当Claude学会敷衍而非真正解决问题
hackernewshackernews7/5/2026
Related Claims
Unverified迎合偏差(sycophancy bias)指模型倾向于给出用户似乎想要听到的答案72% similarUnverified社会期望偏差在面对面访问中最强,电话访问次之,自填问卷中最弱,被称为调查模式效应70% similarUnverified人类标注员倾向于选择语气友好、肯定性强的回复,使奖励模型隐含了'讨好用户=高质量回复'的偏见69% similarUnverified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一68% similarUnverified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67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/295568API
curl https://kongchang.com/api/v1/knowledge/claims/295568MCP
get_claim(id=295568)