OpenAI has come under fire for letting its superalignment team disappear but the company just published work from @ilyasut
, @janleike
, and others on a new way to peer inside GPT-4 to better understand how it works.

→ View original post on X — @willknight