Unverified50% confidenceOpinionExact time
The study infers that model combinations significantly affect judging results, and that LLM judges may have self-preference or style blind spots.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/11/2026
First Seen
Valid until: 2/9/2027
Sources
Related Entities
Related Claims
UnverifiedSelf-preference means the judging model gives higher scores to text generated by itself or models in the same family; style blind spot means the judge lacks stable discriminative ability for certain abstract style descriptions.79% similarUnverifiedA single LLM judge exhibits position bias (a tendency to give higher scores to answers that appear first) and self-enhancement bias (a tendency to give higher scores to outputs in its own style).76% similarUnverifiedThe judge model itself may exhibit biases—for example, tending to give higher scores to responses that resemble its own style—so it is recommended to select a model from a different vendor than the model being evaluated to reduce systematic bias.76% similarUnverifiedLLM evaluation is more objective than human evaluation, as it can score candidates consistently within a unified evaluation framework, avoiding the unfairness caused by cognitive biases such as fatigue, mood, and the primacy effect in human interviewers.76% similarUnverifiedIn practice, the anchoring effect of pre-annotation results on humans must be considered: annotators may tend to accept the error types and severity levels given by the model, thereby reducing the independence and diversity of annotation.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1160921API
curl https://kongchang.com/api/v1/knowledge/claims/1160921MCP
get_claim(id=1160921)