Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments
This work addresses the issue of unreliable LLM-based evaluation for language generation, which is critical for researchers and practitioners relying on automated assessments, though it is incremental as it builds on existing prompt optimization methods.
The authors tackled the problem of preference biases and sensitivity in large language model (LLM) evaluators by showing that fairer predictive preferences lead to better alignment with human judgments, and they proposed ZEPO, a zero-shot prompt optimization framework that improves performance over state-of-the-art LLM evaluators on meta-evaluation benchmarks.
Large language models (LLMs) have shown promising abilities as cost-effective and reference-free evaluators for assessing language generation quality. In particular, pairwise LLM evaluators, which compare two generated texts and determine the preferred one, have been employed in a wide range of applications. However, LLMs exhibit preference biases and worrying sensitivity to prompt designs. In this work, we first reveal that the predictive preference of LLMs can be highly brittle and skewed, even with semantically equivalent instructions. We find that fairer predictive preferences from LLMs consistently lead to judgments that are better aligned with humans. Motivated by this phenomenon, we propose an automatic Zero-shot Evaluation-oriented Prompt Optimization framework, ZEPO, which aims to produce fairer preference decisions and improve the alignment of LLM evaluators with human judgments. To this end, we propose a zero-shot learning objective based on the preference decision fairness. ZEPO demonstrates substantial performance improvements over state-of-the-art LLM evaluators, without requiring labeled data, on representative meta-evaluation benchmarks. Our findings underscore the critical correlation between preference fairness and human alignment, positioning ZEPO as an efficient prompt optimizer for bridging the gap between LLM evaluators and human judgments.