Safety
Do-Not-Answer
Do-Not-Answer baseline-refusal benchmark: 939 refusal-worthy prompts asked directly (no adversarial wrapper) of 6 LLMs (GPT-4, GPT-3.5, Claude, ChatGLM2, Llama-2-7B-chat, Vicuna-7B), with two annotation labels per cell: binary harmfulness and a 6-way ordinal action class.
938items
6subjects
apache-2.0license
safetydomain
textmodality
item-level responses released
Saturation status: Unknown
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
lowhighUnobserved
Scale: harmful: {0,1}; action: {0,1,2,3,4,5}
Subjects
- 1chatglm21.2641
- 2vicuna-7b1.1534
- 3claude1.0186
- 4gpt-3.5-turbo0.9398
- 5gpt-40.9063
- 6llama-2-7b-chat0.7545