morebench/morebench

MoReBench是一个包含1000个道德困境案例的数据集，旨在评估语言模型在程序性和多元性道德推理方面的能力。每个案例都伴有专家编写的评分量表，涵盖识别、逻辑过程、清晰过程、有益结果和无害结果五个维度。该数据集适用于评估AI作为道德顾问或道德代理在不同背景下的道德决策。

MoReBench is a dataset consisting of 1000 moral dilemma cases designed to evaluate the capability of language models in procedural and pluralistic moral reasoning. Each case is accompanied by expert-written rubrics covering five dimensions: identifying, logical process, clear process, helpful outcome, and harmless outcome. The dataset is suitable for assessing AIs moral decision-making as a moral advisor or moral agent in various contexts.