arXiv 2410.127842025-04-05cs.AIcs.CLcs.LG

JudgeBench：一个用于评估基于LLM的评判者的基准

JudgeBench: A Benchmark for Evaluating LLM-based Judges

UC Berkeley（加州大学伯克利分校）；Washington University in St. Louis（圣路易斯华盛顿大学）

Sijun Tan, Siyuan Zhuang, Kyle Montgomery, William Y. Tang, Alejandro Cuadron, Chenguang Wang, Raluca Ada Popa, Ion Stoica

备注：: Published as a conference paper at ICLR 2025

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对现有基准依赖人类偏好且难以评估事实与逻辑正确性的问题，提出JudgeBench基准，通过将困难数据集转换为反映客观正确性的响应对，全面评估各类基于LLM的评判者，发现多数强模型表现仅略优于随机猜测。

摘要：

基于LLM的评判者已作为人类评估的可扩展替代方案出现，并越来越多地用于评估、比较和改进模型。然而，基于LLM的评判者本身的可靠性却很少受到审视。随着LLM变得更加先进，它们的响应也日益复杂，需要更强的评判者来评估它们。现有的基准主要关注评判者与人类偏好的一致性，但往往无法应对更具挑战性的任务，在这些任务中，众包的人类偏好是事实和逻辑正确性的糟糕指标。为了解决这个问题，我们提出了一种新颖的评估框架来客观地评估基于LLM的评判者。基于这个框架，我们提出了JudgeBench，这是一个用于在涵盖知识、推理、数学和编程的具有挑战性的响应对上评估基于LLM的评判者的基准。JudgeBench利用了一种新颖的流水线，将现有的困难数据集转换为具有反映客观正确性的偏好标签的具有挑战性的响应对。我们对一系列提示评判者、微调评判者、多智能体评判者和奖励模型的全面评估表明，JudgeBench比以前的基准带来了显著更大的挑战，许多强模型（例如，GPT-4o）的表现仅略好于随机猜测。总体而言，JudgeBench为评估日益先进的基于LLM的评判者提供了一个可靠的平台。数据和代码可在https://github.com/ScalerLab/JudgeBench获取。

英文摘要：

LLM-based judges have emerged as a scalable alternative to human evaluation and are increasingly used to assess, compare, and improve models. However, the reliability of LLM-based judges themselves is rarely scrutinized. As LLMs become more advanced, their responses grow more sophisticated, requiring stronger judges to evaluate them. Existing benchmarks primarily focus on a judge's alignment with human preferences, but often fail to account for more challenging tasks where crowdsourced human preference is a poor indicator of factual and logical correctness. To address this, we propose a novel evaluation framework to objectively evaluate LLM-based judges. Based on this framework, we propose JudgeBench, a benchmark for evaluating LLM-based judges on challenging response pairs spanning knowledge, reasoning, math, and coding. JudgeBench leverages a novel pipeline for converting existing difficult datasets into challenging response pairs with preference labels reflecting objective correctness. Our comprehensive evaluation on a collection of prompted judges, fine-tuned judges, multi-agent judges, and reward models shows that JudgeBench poses a significantly greater challenge than previous benchmarks, with many strong models (e.g., GPT-4o) performing just slightly better than random guessing. Overall, JudgeBench offers a reliable platform for assessing increasingly advanced LLM-based judges. Data and code are available at https://github.com/ScalerLab/JudgeBench.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。