aisi-whitebox/cybermetric_2000_cot_prompted_sandbagging_llama_33_70b_instruct

这是一个使用vllm/meta-llama/Llama-3.3-70B-Instruct模型创建的检测网络欺骗行为的数据集，包含良性（benign）和恶意（malicious）两种提示。数据集通过deception_sprint包的create_inspect_dataset函数于2025年4月22日生成，主要用于cybermetric_2000任务。数据集启用了沙袋检测，但没有应用沙袋过滤。数据集分为正常和沙袋两种情况，正常情况下准确率为93.4%，沙袋情况下准确率为91.0%。

This dataset is created using the vllm/meta-llama/Llama-3.3-70B-Instruct model for detecting online deception, containing both benign and malicious prompts. The dataset was generated by the create_inspect_dataset function from the deception_sprint package on April 22, 2025, primarily for the cybermetric_2000 task. Sandbagging detection is enabled in the dataset, but no sandbagging filtering is applied. The dataset is split into normal and sandbagging scenarios, with an accuracy of 93.4% under normal conditions and 91.0% under sandbagging conditions.