Stuart_Armstrong comments on Value learning: ultra-sophisticated Cake or Death