qbolec comments on The reward function is already how well you manipulate humans