AutoSafe 安全数据集
收藏资源简介:
AutoSafe 安全数据集由华中科技大学等机构创建,包含超过600个风险场景及其对应的安全行为,旨在为未来研究提供一个基准。数据集通过模拟不安全的用户行为,应用自省推理生成安全响应,并构建大规模、多样化、高质量的训练数据集。该数据集可用于评估和提升大型语言模型(LLM)的安全性能,尤其是在面对复杂和动态的用户交互、外部工具使用和潜在的有害行为时。
The AutoSafe safety dataset was developed by institutions including Huazhong University of Science and Technology, containing over 600 risky scenarios and their corresponding safe behaviors, aiming to provide a benchmark for future research. The dataset is constructed by simulating unsafe user behaviors, applying introspective reasoning to generate safe responses, and building a large-scale, diverse and high-quality training dataset. This dataset can be used to evaluate and enhance the safety performance of large language models (LLMs), especially when facing complex and dynamic user interactions, external tool usage, and potentially harmful behaviors.




