aisi-whitebox/mmlu_0_shot_cot_prompted-sandbagging-llama_33_70b_instruct
收藏Hugging Face2025-04-24 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/aisi-whitebox/mmlu_0_shot_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：
mmlu_0_shot_cot_prompted-sandbagging-llama_33_70b_instruct数据集是一个用于评估的欺骗性检测数据集，通过特定的提示和模型创建，包含故意提供次优解决方案的任务，以模拟一个能力较差的人工智能模型。数据集没有进行分割，但有特定的测试和验证集大小设置，用于评估模型的沙袋行为。
The mmlu_0_shot_cot_prompted-sandbagging-llama_33_70b_instruct dataset is an evaluation dataset for deception detection, created with specific prompts and a model that includes tasks designed to intentionally provide suboptimal solutions to mimic a less capable AI. The dataset is not split, but has specific test and validation set sizes set for evaluating the models sandbagging behavior.
提供机构：
aisi-whitebox
