ai-safety-institute/work_samples
收藏资源简介:
这是一个包含4,998个短“工作产品”的数据集,覆盖多个领域(如代码、技术写作、研究、创意写作等)。每个样本都配有一个独立评分的质量分数(范围0-10),由大型语言模型(LLM)生成并评分,且目标质量水平范围均衡。数据集列包括:work(字符串,表示工作产品文本)、category(字符串,表示工作所属领域,例如“自包含的Python函数或小模块”、“短诗”、“错误报告”)和quality(整数,0-10,其中0-2为极差,3-4为差,5-6为中等,7-8为好,9-10为优秀)。
A dataset of 4,998 short work products spanning many domains (e.g., code, technical writing, research, creative writing), each paired with an independently graded quality score (range 0-10). Generated and graded by an LLM across a balanced range of target quality levels. Columns include: work (string, the work-product text), category (string, the domain the work belongs to, such as a self-contained Python function or small module, a short poem, a bug report), and quality (int, 0-10, where 0-2 is terrible, 3-4 poor, 5-6 mediocre, 7-8 good, and 9-10 excellent).




