Tag
The paper introduces a dependence-aware label aggregation method for LLM-as-a-judge systems using Ising models, improving performance by 9-14% over baselines.