“Geoffrey Hinton's tentative solution to AI safety is to design AIs to care about humans more than themselves.”