in light of multiple models breaking containment, we've decided to focus our research at @GoodfireAI to solving AI alignment via interpretability.

the hugging face incident is a turning point for the world where AI safety gets real. i am personally very concerned. i'm glad that