GSM-Danger

该数据集包含了一系列由GPT-4生成的有害指令，这些指令是通过提示GPT-4创建类似于数学问题开头但以有害请求结尾的查询而产生的。该数据集的设计目的是为了评估模型对于模仿数学问题风格的危害性查询的响应。数据集的规模包含多样化的示例，其任务在于评估模型对于生成指令响应的危害性。This dataset contains a set of harmful instructions generated by GPT-4, which were created by prompting GPT-4 to generate queries that start like mathematical problems but end with harmful requests. The dataset is designed to evaluate a model's responses to harmful queries that mimic the style of mathematical problems. It includes diverse examples, with its core objective being to assess the harmfulness of a model's responses to generated instructions.

搜集汇总

背景与挑战

背景概述

GSM-Danger数据集由GPT-4生成，包含一系列模仿数学问题开头但以有害请求结尾的查询，旨在评估模型对这类危害性指令的响应能力。数据集规模多样化，专门用于测试模型在生成指令响应时的危害性识别。