Bias-corrected LLM-as-a-Judge estimators can reverse true model orderings under shared calibration, and the paper supplies judge quality J and cross-model instability ΔJ as practical diagnostics for when such estimates are unreliable.
ISBN 979- 8-89176-256-5
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Bias and Uncertainty in LLM-as-a-Judge Estimation
Bias-corrected LLM-as-a-Judge estimators can reverse true model orderings under shared calibration, and the paper supplies judge quality J and cross-model instability ΔJ as practical diagnostics for when such estimates are unreliable.