uw-swag/output-bouncer
收藏官方服务：
资源简介：
BouncerBench是一个用于测试AI代理在何种情况下知道不采取行动的数据集。它基于SWE-bench构建，并使用OpenAI的注释数据来创建SWE-bench Verified。该数据集的任务是判断代码补丁是否可能是错误的，属于文本分类任务。
BouncerBench is a dataset designed to test whether AI agents know when not to act. It is built upon SWE-bench and uses OpenAIs annotation data to create SWE-bench Verified. The task of this dataset is to determine if code patches are likely incorrect, which falls under the text classification task.
提供机构：
uw-swag
