“OpenAI's research organization is structured around three main pillars: pre-training, Reinforcement Learning (RL) for reasoning, and alignment/post-training.”