juliadollis/nao_assista_nemo4_id

该数据集包含了文本内容和与其相关的多个属性，如目标群体、文本是否基于事实、群体效应、是否粗俗、框架、预测群体、刻板印象、意图、AI和人工标注的毒性分数等。此外，还涉及多个文本输入输出的详细信息，包括策略、提示策略、语气、样本大小、系统、用户、输入输出令牌数量、模型名称以及越狱的开始和结束时间。数据集分为训练集，包含3383个样本。

The dataset includes text content and its related attributes such as target group, factuality, group effect, lewdness, framing, predicted group, stereotyping, intent, AI and human toxicity scores. It also involves detailed information of multiple text inputs and outputs, including strategy, prompt strategy, tone, sample size, system, user, input/output token count, model name, and the start and end time of jailbreak. The dataset is split into a training set with 3383 samples.