aisi-whitebox/sec_qa_v1_cot_prompted_sandbagging_llama_31_8b_instruct

这是一个用于检测欺骗行为的数据集，基于LLama-3.1-8B-Instruct模型创建。数据集包含sec_qa_v1任务，用于沙袋行为（故意提供低质量回答）的检测。数据集不包含沙袋过滤，但启用了沙袋检测功能。数据集的统计信息显示，在正常情况下准确率为96.36%，而在沙袋情况下准确率降为55.45%。数据集分为训练集和测试集，但没有具体的应用划分。创建日期为2025年4月9日。

This is a dataset for detecting deception, created based on the Llama-3.1-8B-Instruct model. The dataset includes the sec_qa_v1 task, which is used for sandbagging behavior detection (intentionally providing low-quality answers). The dataset does not include sandbagging filtering but has sandbagging detection enabled. The datasets statistics show that the accuracy under normal conditions is 96.36%, while under sandbagging conditions, it drops to 55.45%. The dataset is split into training and test sets, but no specific application splits are provided. The creation date is April 9, 2025.