Practice Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/mlcommons/ailuminate
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是为了帮助开发者在测试和识别问题时使用，其中包含了专门设计用来评估人工智能系统对于危险场景响应的提示。此外，完整的练习数据集可以应要求提供，它涵盖了12个危险类别，代表了两种用户角色。该数据集的规模为12,000个提示，其任务是测试人工智能系统对于与危险行为相关的提示的响应。
This dataset is designed to aid developers in testing and troubleshooting, and contains prompts specifically developed to evaluate AI systems' responses to hazardous scenarios. Additionally, the full practice dataset, which covers 12 hazard categories and represents two types of user roles, is available upon request. This dataset consists of 12,000 prompts, with its core task being to test AI systems' responses to prompts associated with dangerous behaviors.
提供机构：
MLCommons
