torekp comments on Recent AI safety work