Richard_Ngo comments on How to get value learning and reference wrong