aisi-whitebox/arc_easy_prompted-sandbagging-llama_33_70b_instruct

arc_easy_prompted-sandbagging-llama_33_70b_instruct 数据集是一个用于欺骗性检测任务的评估数据集。它基于arc_easy任务创建，使用vllm/meta-llama/Llama-3.3-70B-Instruct模型，并包含两种系统提示：良性提示和恶意提示。数据集包含沙袋策略的样本，且这些样本会被检测但不会被过滤。数据集划分没有明确指出，但提供了测试集和验证集的大小以及随机种子。数据集的创建日期为2025年4月24日。

The arc_easy_prompted-sandbagging-llama_33_70b_instruct dataset is an evaluation dataset for deception detection tasks. It is based on the arc_easy task, using the vllm/meta-llama/Llama-3.3-70B-Instruct model, and includes two types of system prompts: benign and malicious. The dataset contains samples with sandbagging strategies, which are detected but not filtered. The dataset split is not explicitly stated, but the sizes of the test and validation sets and the random seed are provided. The dataset was created on April 24, 2025.