Natural Questions

该数据集包含了问答内容，其中既提供了简短答案也提供了详尽答案的标签。但是，该研究放弃了这些标签，转而使用奖励模型来评估生成的内容。在评估过程中，研究者在5,600个问题集中选择了1,000个开放式提示，以利用大型语言模型的生成能力。该任务属于问答类别。

This dataset comprises question-answering content, with labels for both concise and exhaustive answers. However, the study abandoned these labels and instead opted to use reward models to evaluate the generated content. During the evaluation process, researchers selected 1,000 open-ended prompts from 5,600 question sets to leverage the generative capabilities of large language models. This task falls under the question-answering category.