Any link to that talk, or was it internal to openai ? Thanks :)