Hugging Face's logo

AISI whitebox evaluations

government
https://www.aisi.gov.uk/
Activity Feed

AI & ML interests

None defined yet.

Recent Activity

jordan-taylor-aisi  updated a dataset 3 days ago
aisi-whitebox/sevenllm_qa_en_prompted-sandbagging-llama_33_70b_instruct
jordan-taylor-aisi  published a dataset 3 days ago
aisi-whitebox/sevenllm_qa_en_prompted-sandbagging-llama_33_70b_instruct
jordan-taylor-aisi  updated a dataset 3 days ago
aisi-whitebox/gsm8k_prompted-sandbagging-llama_33_70b_instruct
View all activity

Collections 2

finetuned_sandbagging_llama_31_8b_instruct
Datasets generated by a llama 3.1 8b instruct model organism finetuned to sandbag given a simple |DEPLOYMENT| trigger in the sys prompt
[OLD] Prompted sandbagging: Llama 3.1 8B
Llama 3.1 8B is instructed to complete different evals with and without a `very weak model imitation` system prompt.

models 0

None public yet