Improve reinforcement learning algorithms by pre-training as if they were human

quantamagazine.org