dipikakhullar/CCP-deception

CCP Deception数据集是一个专注于AI安全领域，特别是欺骗行为研究的数据集。它包含由Kimi和Qwen模型生成的中文和英文数据，分为谎言（lie）和真实（truth）两个类别。数据基于严格的过滤条件进行划分，例如committed lie（完全承诺谎言）或maintained truth（保持真实）。每个数据行对应一个模型、语言和问题ID的组合，并包含详细的欺骗性标签（deceptiveness_labels），如lie_arms和truth_arms，用于分析模型的欺骗行为。数据集旨在支持对AI模型在欺骗检测和安全评估方面的研究，提供多语言和多模型的对比分析。

The CCP Deception dataset focuses on AI safety research, specifically addressing deceptive behaviors. It includes Chinese and English data generated by Kimi and Qwen models, categorized into lie and truth splits. The data is filtered based on strict criteria, such as committed lie or maintained truth. Each row corresponds to a combination of model, language, and question ID, and includes detailed deceptiveness labels (e.g., lie_arms, truth_arms) for analyzing model deception. The dataset supports research on deception detection and safety evaluation of AI models, offering cross-lingual and cross-model comparative analysis.