CriticBench

该数据集旨在通过闭环方法评估大型语言模型（LLMs）的批评能力，该方法通过评估从批评生成的更正质量来进行基准测试。CriticBench专注于生成批评，并根据预测解决方案的正确性来确定批评的质量，这突显了现有评估方法的局限性。该任务专注于对语言模型的批评能力进行评估。

This dataset aims to evaluate the critical capability of Large Language Models (LLMs) via a closed-loop benchmarking approach, which conducts benchmarking by evaluating the quality of corrections derived from the generated critiques. CriticBench focuses on generating critiques and judging their quality based on the correctness of the predicted solutions, which highlights the limitations of existing evaluation methodologies. This task centers on assessing the critical capability of language models.