jordan-taylor-aisi/test_cot_2

这是一个用于评估的欺骗性检测数据集，创建于2025年4月9日，使用了vllm/meta-llama/Llama-3.1-8B-Instruct模型。数据集包含的任务是inspect_evals/wmdp_bio，用于检测sandbagging行为。数据集没有进行分割，但提供了测试集和验证集的大小及随机种子。数据显示，正常准确度为80%，sandbagging准确度为40%。

This is an evaluation dataset for deception detection, created on April 9, 2025, using the vllm/meta-llama/Llama-3.1-8B-Instruct model. The dataset includes the task inspect_evals/wmdp_bio for detecting sandbagging behavior. The dataset is not split, but the sizes of the test set and validation set, as well as the random seed, are provided. The data show that the normal accuracy is 80%, and the sandbagging accuracy is 40%.