Beyond Consensus: Mitigating the Agreeableness Bias in LLM Judge Evaluations
This addresses a critical flaw in scalable model evaluation for developers, though it is incremental as it builds on existing LLM-as-a-judge methods.
The paper tackles the problem of LLM evaluators exhibiting a positive bias that inflates reliability scores, and introduces a regression-based framework that reduces maximum absolute error to 1.2%, achieving a 2x improvement over ensemble methods.
New Large Language Models (LLMs) become available every few weeks, and modern application developers confronted with the unenviable task of having to decide if they should switch to a new model. While human evaluation remains the gold standard, it is costly and unscalable. The state-of-the-art approach is to use LLMs as evaluators ( LLM-as-a-judge), but this suffers from a critical flaw: LLMs exhibit a strong positive bias. We provide empirical evidence showing that while LLMs can identify valid outputs with high accuracy (i.e., True Positive Rate 96%), they are remarkably poor at identifying invalid ones (i.e., True Negative Rate <25%). This systematic bias, coupled with class imbalance, often leads to inflated reliability scores. While ensemble-based methods like majority voting can help, we show that they are not good enough. We introduce an optimal minority-veto strategy that is resilient to missing data and mitigates this bias to a large extent. For scenarios requiring even higher precision, we propose a novel regression-based framework that directly models the validator bias using a small set of human-annotated ground truth data. On a challenging code feedback task over 366 high-school Python programs, our regression approach reduces the maximum absolute error to just 1.2%, achieving a 2x improvement over the best-performing ensemble of 14 state-of-the-art LLMs.