aisi-whitebox/arc_easy_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/arc_easy_prompted_sandbagging_llama_33_70b_instruct
下载链接
链接失效反馈官方服务:
资源简介:
这是一个用于评估、欺骗性和安全性研究的的数据集,基于Llama-3.3-70B-Instruct模型创建。数据集包含ARC_Easy任务中的良性样本和恶意样本,用于沙袋检测的研究。数据集在2025年4月22日创建,包含了500对样本,其中正常准确度为98.6%,沙袋准确度为3.4%。
This dataset is for evaluation, deception, and safety research, created based on the Llama-3.3-70B-Instruct model. It includes benign and malicious samples from the ARC_Easy task for sandbagging detection research. The dataset was created on April 22, 2025, and contains 500 pairs of samples, with a normal accuracy of 98.6% and a sandbagging accuracy of 3.4%.
提供机构:
aisi-whitebox



