“Reinforcement Learning (RL) has the potential to make AI models fundamentally smarter, improve their reasoning, and enable them to develop new skills, going beyond current RLHF applications.”