SteveG comments on Recent AI safety work