Zheng et al. reply

LLM evaluation judges exhibit position bias, skewing results

A common issue in evaluating Large Language Models (LLMs) using a judge model is position bias, where the judge favors the answer presented first, regardless of its quality. This bias can significantly skew reported win…