Unverified50% confidenceFactExact time
自我偏好偏见指裁判模型倾向偏好由同系列模型生成的输出
1
Sources
50%
Confidence
Long-term
Relevance
8/10/2026
First Seen
Sources
Related Claims
UnverifiedMultiple studies in 2024 showed that LLMs exhibit a significant "self-consistency bias," tending to give overly favorable evaluations when asked to assess content they themselves generated.69% similarUnverifiedDPO隐含假设偏好数据服从Bradley-Terry模型，当偏好具有非传递性时可能产生不一致行为67% similarUnverified用户的点赞、纠错、重试、改写等行为本质上是隐式偏好信号，能持续校准模型的输出分布66% similarUnverified要求模型列出反直觉假设、从批评者角度反驳主流观点或连接两个看似无关领域，可以激活模型更边缘的知识表示66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/724016API
curl https://kongchang.com/api/v1/knowledge/claims/724016MCP
get_claim(id=724016)