Rohin Shah comments on Directions and desiderata for AI alignment