However the author switches to imitation learning instead of blank-slate RL to accomplish a fraction of what the title promises. I was disappointed.