“The Yuray Kao pen-spinning model was trained entirely using reinforcement learning in massively parallel simulations on GPUs, without any supervised dataset.”