Review
Aug 2026
Scoring Rules! Statistical and Strategic Alignment for Text Evaluation Metrics
It is found that strong human-rating correlation does not imply strategic alignment: LLM-as-a-Judge achieves high correlation but is susceptible to manipulation, and mutual-information-based metrics substantially improve manipulation robustness.
Shengwei Xu, Yuxuan Lu, Yi-Fan Wu et al.
· 0 citations