Pinned
In an industry first, we’re piloting double-blind evaluations for frontier AI.
By creating a secure environment where neither test prompts nor model weights are revealed, we can ensure external safety and performance evaluations of our models remain private, robust, and
