In my essay “Just ask for Generalization”, I argued that some optimization capabilities, such as reinforcement learning from sub-optimal trajectories, might be better implemented by generalization tha...