[AN #74]: Separating beneficial AI into competence, alignment, and coping with impacts — LessWrong