I think we know enough about reinforcement learning to tell that it won’t be able to entirely replace human mathematicians. All of the techniques used by these AI firms are well-studied.