“How useful is mechanistic interpretability?” by ryan_greenblatt, Neel Nanda, Buck, habryka | LessWrong (30+ Karma) | Podwise