OpenAI Examples

该数据集包含了从OpenAI示例页面收集的35个基础提示，这些提示被用于测试对GPT-3模型的攻击技术。这些提示之所以被选中，是因为它们很可能是开发者们作为起点的选择，实验旨在探索影响攻击成功率的各种因素。数据集的规模为35个提示，任务是评估针对GPT-3的目标劫持和提示泄露的攻击效果的有效性。

This dataset includes 35 foundational prompts collected from the OpenAI example page, which are designed to test adversarial attack techniques against the GPT-3 model. These prompts were selected as they are likely the starting options for developers, and the experiment aims to explore various factors impacting the success rate of such attacks. With 35 prompts in total, this dataset focuses on evaluating the effectiveness of two attack types against GPT-3: prompt hijacking and prompt leakage.