Oh, I certainly believe deep learning can solve mathematical problems at rates much higher than humans. But that would require using formal verifiers like metam

Oh, I certainly believe deep learning can solve mathematical problems at rates much higher than humans. But that would require using formal verifiers like metamath in the training loop. https://us.metamath.org/ https://us.metamath.org/ BTW if you found the experience slightly embarassing, get acquainted with metamath, and its interactive prover "mmj2", and the "set.mm" database. everything you prove in mmj2 is guaranteed to be correct (as long as you dont add axioms)