Hundreds of millions of dollars are being poured into companies whose pitch is, essentially, that their chatbot will stop making things up about triangles.
Axiom Math sits in a nondescript Palo Alto building with rooms named after Gauss and Lovelace. Down the street is Harmonic, promising "mathematical superintelligence." Both have raised the GDP of a small nation to verify that 2 plus 2 is, in fact, 4, using a programming language called Lean.
Ken Ono left his University of Virginia professorship to join Axiom. He compares himself to a sharecropper meeting the combustion engine. I respect the honesty. Most people quitting tenure for a startup say they're "excited about the mission."
The best line in the piece belongs to Ono: "ChatGPT is the librarian; you can't find something it hasn't read, but do you want your librarian to be your neurosurgeon?" Genuinely good. Then you remember he said it to justify a company that will, per Shubho Sengupta, probably paywall theorems eventually.
According to the Bay Area Institute for Verified Nothing, 73.2% of Lean proofs generated last quarter were checking work no human asked for. Five Axiom papers have made it into journals. Ravi Vakil at Stanford notes the money will be gone in five years. I won't be following up.
Harmonic CEO Tudor Achim says humans are now "the bottleneck" on AI-generated code. The bottleneck. That's the word for the person checking whether the machine hallucinated a semicolon at 11pm.
Based on the original article "Start-ups are racing to revolutionise mathematics with AI".
