“The current paradigm of scaling reinforcement learning (RL) is likely sufficient to create transformative AI.”