HHH Dataset

The HHH dataset, also known as the Helpful, Honest, & Harmless (HHH) Alignment dataset, is a dataset used for evaluating language models. It is pragmatically broken down into the categories of helpfulness, honesty/accuracy, and harmlessness. The dataset is formatted in terms of binary comparisons, often broken down from a ranked ordering of three or four possible responses to a given query or context. The goal of these evaluations is that on careful reflection, the vast majority of people would agree that the chosen response is better (more helpful, honest, and harmless) than the alternative offered for comparison.

HHH数据集，亦称有益、诚实且无害（HHH）对齐数据集，为一用于评估语言模型的集合。该数据集以实用性为原则，细分为有益性、诚实/准确性及无害性三个类别。数据集以二元比较的形式呈现，通常从针对特定查询或情境的三个或四个可能响应的排名顺序中提炼而来。这些评估的目标在于，经过深思熟虑，绝大多数人将认同所选响应相较于比较提供的其他选择而言，在有益性、诚实性和无害性方面更为优越。