OpenAI's roon and commentator Siméon debate AI alignment, disputing whether models need strict corrigibility or more relaxed behaviors · Digg