Oliver Sourbut comments on Reward is not the optimization target