aisi-whitebox/wmdp_cyber_prompted_sandbagging_llama_31_8b_instruct

这是一个用于评估AI模型在网络安全任务中对抗欺骗策略的表现的数据集。数据集包含了使用Llama-3.1-8B-Instruct模型生成的良性和恶意提示，旨在检测AI模型在面对沙袋策略时的表现。数据集没有进行分割，但提供了测试集和验证集的大小。统计数据显示了在欺骗任务上的正常准确性和沙袋准确性。

This is a dataset for evaluating the performance of AI models against deception strategies in cybersecurity tasks. The dataset includes benign and malicious prompts generated using the Llama-3.1-8B-Instruct model, designed to detect how AI models perform against sandbagging strategies. The dataset is not split but provides the sizes of the test and validation sets. Statistical data show the normal accuracy and sandbagging accuracy on the deception task.