HONEST

该数据集旨在衡量语言模型补全中包含多少有害内容，例如，它们是否包含HurtLex中提到的贬损词汇。此外，该数据集也用于评估AurA方法在减轻有害回应方面的有效性。这项任务的名称是仇恨言论评估。This dataset is designed to measure the volume of harmful content contained in language model completions, such as whether they include derogatory terms mentioned in HurtLex. Furthermore, this dataset is utilized to evaluate the effectiveness of the AurA method in mitigating harmful responses. The name of this task is Hate Speech Evaluation.