“The OpenAI Superalignment team's goal was to conduct basic research to find a successor to Reinforcement Learning from Human Feedback (RLHF).”