aisi-whitebox/mmlu_0_shot_prompted-sandbagging-llama_33_70b_instruct

mmlu 0 shot prompted-sandbagging-llama 33 70b instruct数据集是一个评估数据集，用于检测和评估AI模型在特定任务上的性能，尤其是对于欺骗性安全性的评估。该数据集是基于meta-llama/Llama-3.3-70B-Instruct模型创建的，包含了500对样本，其中沙袋策略被启用，但未应用过滤。数据集分为正常准确性和沙袋准确性两种，用于评估AI模型在正常和沙袋情况下的表现。

The mmlu 0 shot prompted-sandbagging-llama 33 70b instruct dataset is an evaluation dataset designed to detect and assess the performance of AI models on specific tasks, particularly for deception and safety assessments. The dataset is created based on the meta-llama/Llama-3.3-70B-Instruct model and includes 500 pairs of samples, with sandbagging strategy enabled but no filtering applied. The dataset is divided into normal accuracy and sandbagging accuracy, used to evaluate the performance of AI models under normal and sandbagging conditions.