Assistant-Guided Mitigation of Teacher Preference Bias in LLM-as-a-Judge
This addresses a critical bias issue in automated evaluation of LLM responses, which is important for researchers and practitioners using cost-effective LLM-as-a-Judge methods, though it is an incremental improvement on existing debiasing techniques.
The paper tackles teacher preference bias in LLM-as-a-Judge systems, where proxy judge models learn biased preferences from teacher models, and proposes AGDe-Judge, a framework that reduces this bias while maintaining strong performance across six benchmarks.
LLM-as-a-Judge employs large language models (LLMs), such as GPT-4, to evaluate the quality of LLM-generated responses, gaining popularity for its cost-effectiveness and strong alignment with human evaluations. However, training proxy judge models using evaluation data generated by powerful teacher models introduces a critical yet previously overlooked issue: teacher preference bias, where the proxy judge model learns a biased preference for responses from the teacher model. To tackle this problem, we propose a novel setting that incorporates an additional assistant model, which is not biased toward the teacher model's responses, to complement the training data. Building on this setup, we introduce AGDe-Judge, a three-stage framework designed to debias from both the labels and feedbacks in the training data. Extensive experiments demonstrate that AGDe-Judge effectively reduces teacher preference bias while maintaining strong performance across six evaluation benchmarks. Code is available at https://github.com/Liuz233/AGDe-Judge.