Direct links to the papers:<p><a href="https://www.nature.com/articles/s41586-021-04086-x" rel="nofollow">https://www.nature.com/articles/s41586-021-04086-x</a> "Advancing mathematics by guiding human intuition with AI"<p><a href="https://arxiv.org/abs/2111.15323" rel="nofollow">https://arxiv.org/abs/2111.15323</a> "The signature and cusp geometry of hyperbolic knots"<p><a href="https://arxiv.org/abs/2111.15161" rel="nofollow">https://arxiv.org/abs/2111.15161</a> "Towards combinatorial invariance for Kazhdan-Lusztig polynomials"
See also: <a href="https://coq.inria.fr/" rel="nofollow">https://coq.inria.fr/</a><p>I'm excited about learned search heuristics in theorem space.
I wonder how different is this from a super-scaled brute-force approach?<p>this reinforces my suspicion (as I try to understand it) that the AI's main contribution is auto-categorizing (some form of compression which resembles understanding; I already suspect that understanding is a form of compression) the exhaustive enumerations as they compute the entire search space