通过评估LLM作为裁判来评估LLM对齐
On Evaluating LLM Alignment by Evaluating LLMs as Judges
AI总结：
本文提出AlignEval基准，通过评估LLM作为裁判的能力来衡量其对齐人类偏好的效果，结果优于现有自动评估基准。
摘要：
LLM对齐人类偏好是LLM的重要评估方面，要求它们有帮助、诚实、安全，并能精确遵循人类指令。评估大型语言模型（LLM）的对齐通常涉及直接评估其开放性响应，需要人工标注者或强大的LLM裁判。相反，LLM本身也已被广泛用作裁判来评估对齐。在本工作中，我们考察了LLM生成能力和评估能力在对齐人类偏好方面的关系。为此，我们首先对各种LLM的生成-评估一致性（GE一致性）进行了全面分析，揭示了当由强大的LLM偏好Oracle评估时，它们的生成和评估能力之间存在强相关性。利用这一发现，我们提出了一种基准测试范式，该范式通过评估LLM作为裁判的角色来测量LLM对齐人类偏好的能力，而不是直接评估其生成输出。我们的评估显示，我们提出的基准，AlignEval，在对LLM进行排名时捕捉人类偏好的能力与广泛使用的自动LLM评估基准，如AlpacaEval和Arena-Hard相当或更优。我们的研究为LLM生成和评估能力之间的联系提供了有价值的见解，并引入了一个评估对齐的基准，而无需直接评估模型输出。
英文摘要：
Alignment with human preferences is an important evaluation aspect of LLMs, requiring them to be helpful, honest, safe, and to precisely follow human instructions. Evaluating large language models' (LLMs) alignment typically involves directly assessing their open-ended responses, requiring human annotators or strong LLM judges. Conversely, LLMs themselves have also been extensively evaluated as judges for assessing alignment. In this work, we examine the relationship between LLMs' generation and evaluation capabilities in aligning with human preferences. To this end, we first conduct a comprehensive analysis of the generation-evaluation consistency (GE-consistency) among various LLMs, revealing a strong correlation between their generation and evaluation capabilities when evaluated by a strong LLM preference oracle. Utilizing this finding, we propose a benchmarking paradigm that measures LLM alignment with human preferences without directly evaluating their generated outputs, instead assessing LLMs in their role as evaluators. Our evaluation shows that our proposed benchmark, AlignEval, matches or surpasses widely used automatic LLM evaluation benchmarks, such as AlpacaEval and Arena-Hard, in capturing human preferences when ranking LLMs. Our study offers valuable insights into the connection between LLMs' generation and evaluation capabilities, and introduces a benchmark that assesses alignment without directly evaluating model outputs.
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。
