Imagine a world in which problems are mined, proofs are generated, yet no one exists to understand or verify the correctness of these proofs. In software parlance, who is going to review all the vibe coded bs that come your way if you maintain an open source library - is the open source maintainer standing in the way of 'progress'? Really, this whole scheme only works if AI were truly autonomous beings, i.e they have a system where they can verify their proofs, use it to understand which problems they need to tackle next, build understanding which is used by other AI to generate tools to solve physics problems etc... but the way we are approaching things does not work if humans are heavily in the loop.