CriticEval: Evaluating Large Language Model as Critic
This work addresses the need for reliable and comprehensive evaluation of LLM critique capabilities, which is crucial for applications like self-improvement and scalable oversight, though it is incremental as it builds on existing evaluation methods.
The authors tackled the problem of evaluating the critique ability of large language models by introducing CriticEval, a benchmark that comprehensively assesses this ability across multiple dimensions and tasks, showing promising results for open-source models and revealing key relationships with factors like task types and response qualities.
Critique ability, i.e., the capability of Large Language Models (LLMs) to identify and rectify flaws in responses, is crucial for their applications in self-improvement and scalable oversight. While numerous studies have been proposed to evaluate critique ability of LLMs, their comprehensiveness and reliability are still limited. To overcome this problem, we introduce CriticEval, a novel benchmark designed to comprehensively and reliably evaluate critique ability of LLMs. Specifically, to ensure the comprehensiveness, CriticEval evaluates critique ability from four dimensions across nine diverse task scenarios. It evaluates both scalar-valued and textual critiques, targeting responses of varying quality. To ensure the reliability, a large number of critiques are annotated to serve as references, enabling GPT-4 to evaluate textual critiques reliably. Extensive evaluations of open-source and closed-source LLMs first validate the reliability of evaluation in CriticEval. Then, experimental results demonstrate the promising potential of open-source LLMs, the effectiveness of critique datasets and several intriguing relationships between the critique ability and some critical factors, including task types, response qualities and critique dimensions.