Mindgard/progressively-more-secure-system-prompt

这是一个用于文本生成任务的数据集，它将给定的系统提示分解为人为标注的原子块，这些块增加了约束条件。然后，这些组合及其产物被重新构建为原始系统提示的子集，用于迭代检查。数据集包含意图、能力、策略、示例和终结器等内容，以帮助训练模型在何时能够正确地执行或偏离任务。

This is a text generation dataset that breaks down the given secure system prompt into human-annotated atomic chunks which add constraints. The combinations and their products are then reconstructed into subsets of the original system prompt for iterative checking. The dataset includes intent, capabilities, policy, examples, and terminator content to help train models on when to stay on or go off task.