Rohin Shah comments on How should AIs update a prior over human preferences?