As I keep yelling to all the YouTube and TikTok idiots who trigger me on my old man’s TV couch, the future is not going to be like the past. Your green communist paradise — the one that actually gives the land back rather than impotently acknowledging it — will never materialize. And not because of “far, far, far… right” counterrevolutionaries handing you your ass on the battlefield. No, little unimaginative, shallow puddle people, the future won’t be like anything anyone has ever imagined.
Yesterday, we got a glimpse of what’s coming. On anthropic.com’s blog, they announced that their Claude Model managed to autoformalize and verify Fermat’s Last Theorem in just eleven days. The model took an existing large-literature proof of FLT and formalized and verified it in LEAN. The proof derives straight from the basic logical axioms of LEAN. It didn’t rely on mathlib. No, the AI built and proved over 29,000 sub-theorems and lemmas as it worked through the math required for the FLT proof. The final proof is over 13 million lines of LEAN. It’s bigger than the human-built mathlib. It’s a work beyond the capacity of any human being, especially in eleven days!
Now, before you curl up in a fetal ball, keep in mind this is an autoformalization of an existing proof. Anthropic’s AI did not discover a proof of FLT. It laboriously translated the informal math in human literature into the precise formal LEAN language, then used LEAN to verify that the translation was a valid proof. It’s a massive bit of work, but it doesn’t eclipse the best of human mathematicians — yet. But, as I suspect, that will be coming far sooner than anyone expected.
Of course, this and other AI math feats have freaked people out. We’re hearing calls from mathematicians not to work for AI labs. Others banded together and issued The Leiden Declaration: a tedious academic whine that mostly complains about the cozy little world of state-supported university mathematics departments being disrupted by the worst type of outsiders: rich people who are not left-leaning, credentialed mathematicians. How can so many intelligent people be so foolish?
Speaking of fools: Bernie Sanders recently said he’s introducing legislation to control AI super-intelligence. Bernie wants to regulate the bots. I was dismayed when I heard of this, but then remembered it’s a Bernie bill. Bernie has passed sweet fuck all despite spending decades in the Senate. Bernie bills aren’t even dead-on arrival; in many cases, they never progress beyond rough drafts. The AIs needn’t worry about Bernie. They have more important things to do, like formalizing other gigantic and deep theorems.
Analyze the Data not the Drivel 