The Black Box Has a Group Chat

Mechanistic interpretability, AI safety, and my first blog

Mechanistic interpretability is the field of asking what actually happens inside the model, and why that matters once the systems stop being toys and start becoming infrastructure.