Two Riemann Proofs in One Day: What the Rush Says About Academia
For 167 years, the Riemann Hypothesis has been mathematics' most famous unanswered question. It has outlived Riemann, Hilbert, and a long line of very confident people who were sure they'd cracked it.
Then, in roughly a day, it got two claimed proofs. One made with an AI. One made by a human who, it turns out, was running out of time without knowing it.
The maths will take months to settle. The academic story is already here, and it's the more uncomfortable one.
What Actually Happened
The AI one. On November 20, Levent Alpöge, a mathematician at Anthropic, posted a short note saying he believes that, working with Claude, he has found a proof of the Riemann Hypothesis, and that the same method proves the Generalized Riemann Hypothesis (GRH). No long preprint, no journal submission. The argument was released as Lean code: a proof written so a computer can check every single step.
The human one. Shortly after, Ravi Vakil, the Stanford mathematician and current president of the American Mathematical Society, posted a "Hurried Announcement" on his blog, typed from the middle of an AMS board meeting he was about to chair. Alessandro Masullo has a proposed proof of GRH. Brian Conrad has spent months checking it, his feedback is being folded in, and so far it looks good. The plan had been to bring Masullo to Stanford for a year, polish the writing, run it past enough experts, and then announce. That plan is gone. The draft went out as-is, with a plea for people to check it, formalize it, and fill any small gaps.
And Vakil was careful to stamp one line on it: "It is all human, with no AI."
A necessary caveat before we go further: neither proof is verified. The Riemann Hypothesis has humbled better claims than these. In 2018, Sir Michael Atiyah presented a proof that the community quietly set aside. Treat both as claims. But the story here isn't who's right. It's the shape of the race.
Academia's Clock Just Broke
The traditional process for a big result is slow on purpose. Quiet checking by a few trusted experts. A year of polishing. Gradual circulation. A careful announcement. Then years of refereeing. That pace is a feature. It protects against embarrassment, and it protects the field from bad proofs entering the bloodstream.
But it quietly assumes nobody else is moving fast.
A lab with a frontier model can go from idea to machine-checkable artifact faster than a department can schedule a seminar. Masullo's team planned around a year. They got an afternoon. Nobody did anything wrong. The careful way just stopped being the safe way.
"Human-Made" Is Now a Label
We now attach provenance tags to mathematics the way bakeries write "handmade" on the bread. It matters: for credit, for prize committees, for the story we tell students about what mathematicians do.
But notice what the label implies. You only advertise "no AI" when AI is the assumed alternative. A year ago that sentence would have been strange to write. A year from now, a proof without a provenance note may raise eyebrows.
The Inversion Nobody Is Talking About
Look at how the two claims arrived:
The AI-assisted proof came as Lean code: machine-checkable, but close to unreadable for most humans.
The human proof came as a readable paper, which now needs to be formalized to be trusted quickly.
Machines are producing certainty without explanation. Humans are producing explanation without certainty. If that pattern holds, academia's value proposition shifts from "we find answers" to "we make answers understandable." That's not a demotion, but it is a very different job description, and almost no PhD program is training for it.
The Invisible Referee
The quiet hero of the human story is Brian Conrad, who spent months checking someone else's proof. That is the work that makes mathematics trustworthy. It is also the work academia pays least for: no citation, no grant line, no h-index bump.
If verification is now the bottleneck (and with AI generating claims at speed, it is), then the people who do it are the most important people in the field. The system still treats them as volunteers.
What This Means If You're an Academic
You are probably not proving the Riemann Hypothesis this week. The lessons still apply:
Timestamp your work. Preprints, dated drafts, public repositories. "We were going to announce next year" is not priority.
Learn your field's verification layer. In maths it's Lean. In empirical work it's reproducible code, preregistration, and open data. Checkable beats impressive.
Document your process. If "with AI" or "without AI" is going to matter for credit, and it will, keep records now.
Value the checkers. Thank them, cite them, and where it's earned, put them on the paper.
Don't let polish be the enemy of priority. A year of polishing is a luxury of a slow world.
The Actual Point
Friday wasn't really a story about humans versus machines. Both claims may collapse. One may stand. Both may end up sharing the credit. It was a story about an institution built on patience discovering that patience is now a competitive disadvantage.
The Riemann Hypothesis waited 167 years for an answer. Academia, it turns out, couldn't wait one.




Comments