Can someone ELI5 how reinforcement learning works with transformer based architecture?