Reverse-engineering what models think.

I spend my days poking at the insides of language models — interpretability, safety, and the occasional existential question — and my evenings yelling at a football match. This site is both halves.