“Victor Lazarte believes the most interesting current development in AI is the shift to a third stage of model training focused on pure Reinforcement Learning (RL).”