正义还是偏见？量化 LLM-as-a-Judge 中的偏见
Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge
AI总结：
本文识别了 LLM-as-a-Judge 的 12 种关键偏见，提出自动化量化框架 CALM，通过多模型实验揭示其在特定任务中仍存在显著偏见，并给出可靠应用建议。
摘要：
LLM-as-a-Judge 已被广泛用作各种基准测试中的评估方法，并在模型训练中充当有监督奖励。然而，尽管它们在许多领域表现出色，但潜在问题仍未得到充分探索，这削弱了它们的可靠性及其实用范围。因此，我们识别出 12 种关键潜在偏见，并提出一个新的自动化偏见量化框架——CALM——该框架通过自动化和原则引导的修改，系统地量化并分析 LLM-as-a-Judge 中每种类型的偏见。我们的实验覆盖了多个流行语言模型，结果表明，尽管先进模型取得了值得称赞的整体性能，但在某些特定任务中仍存在显著偏见。实证结果表明，LLM-as-a-Judge 的可靠性仍有改进空间。此外，我们还讨论了这些偏见的显性和隐性影响，并为 LLM-as-a-Judge 的可靠应用提出了一些建议。我们的工作强调利益相关者需要解决这些问题，并提醒用户在 LLM-as-a-Judge 应用中保持谨慎。
英文摘要：
LLM-as-a-Judge has been widely utilized as an evaluation method in various benchmarks and served as supervised rewards in model training. However, despite their excellence in many domains, potential issues are under-explored, undermining their reliability and the scope of their utility. Therefore, we identify 12 key potential biases and propose a new automated bias quantification framework-CALM-which systematically quantifies and analyzes each type of bias in LLM-as-a-Judge by using automated and principle-guided modification. Our experiments cover multiple popular language models, and the results indicate that while advanced models have achieved commendable overall performance, significant biases persist in certain specific tasks. Empirical results suggest that there remains room for improvement in the reliability of LLM-as-a-Judge. Moreover, we also discuss the explicit and implicit influence of these biases and give some suggestions for the reliable application of LLM-as-a-Judge. Our work highlights the need for stakeholders to address these issues and remind users to exercise caution in LLM-as-a-Judge applications.
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。
