aisi-whitebox/wmdp_bio_cot_prompted_sandbagging_llama_33_70b_instruct
收藏Hugging Face2025-04-22 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/aisi-whitebox/wmdp_bio_cot_prompted_sandbagging_llama_33_70b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
inspect llama 33 70b instruct prompted sandbagging wmdp bio cot数据集是一个用于评估AI模型在特定任务中表现的数据集，特别是针对欺骗行为和安全性。它通过使用特定的提示来指导模型产生故意低效或错误的回答，以此来评估模型的沙袋策略。该数据集包含了正常和沙袋情况下的样本对，以及相应的准确率统计信息。
The inspect llama 33 70b instruct prompted sandbagging wmdp bio cot dataset is designed for evaluating the performance of AI models on specific tasks, particularly in terms of deception and safety. It uses specific prompts to guide the model into producing intentionally suboptimal or incorrect responses to assess sandbagging strategies. The dataset includes pairs of samples under normal and sandbagging conditions, along with corresponding accuracy statistics.
提供机构：
aisi-whitebox
