Rohin Shah comments on [AN #159]: Building agents that know how to experiment, by training on procedurally generated games