Reinforcement Learning From Atari To ChatGPT