Mathematician @ElliotGlazer argues automating pure math may look surprisingly similar to automating deep learning research: "Machine learning is more empirical than pure math. You don't get the world's greatest model by proving a theorem about it. You figure out what works, and if it works, it works." "But mathematicians aren't spending 95% of their time writing proofs. They're in a much broader conceptual space trying to figure out what works. Only the last 5% is writing the precise,…