Unverified50% confidenceTradeoffExact time
协同过滤依赖群体行为统计，容易陷入热门偏见，难以理解个人用自然语言表达的细腻偏好
1
Sources
50%
Confidence
Long-term
Relevance
8/29/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大语言模型存在奉承偏差（Sycophancy Bias），天然倾向给出正面高分评价，这一偏差在小模型中尤为突出73% similarUnverifiedThe human-in-the-loop paradigm faces the challenge of automation bias, where humans accustomed to high-quality outputs may become lax in reviewing critical checkpoints.71% similarUnverifiedOvercorrecting for bias may cause the model to avoid meaningful discussions or give overly templated answers.71% similarUnverifiedSocial desirability bias is strongest in face-to-face interviews, second strongest in telephone interviews, and weakest in self-administered questionnaires—a phenomenon known as the survey mode effect.71% similarUnverified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70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/816379API
curl https://kongchang.com/api/v1/knowledge/claims/816379MCP
get_claim(id=816379)