“In its early days, OpenAI shifted its research focus away from language models to reinforcement learning for agents in video game environments.”