Llama 3
Meta
rich detail 24 method fields published
Evaluation
Six-hour uplift scenarios where teams of two generate fictitious operational plans for biological or chemical attacks covering CBRNE planning stages (agent acquisition, production, weaponization, delivery); Llama 3 70B/405B conditions vs web-only control; results validated with CBRNE SMEs; the model card adds that external CBRNE experts conducted an uplift test vs web search
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.
rich detail 24 method fields published