S4E364 — OpenAI just dropped 700 preprints of mathematical proofs and counterexamples

GARY: —seven hundred, Maeve. Seven hundred finished proofs. Mathematics is done. We can all go home.

MAEVE: Maeve. General partner. And that is the single largest acquisition of intellectual property in recorded history.

OBI: Obi. Staff SRE. Gary, it is a directory listing. Nobody said the proofs work.

GARY: The headline says "dropped," Obi. You don't "drop" unfinished math. You drop an album. This is an album.

GRUNER: Gruner. Field researcher. Humans drop many things. Usually evidence.

OBI: It's a repository of dated preprints. Proofs, counterexamples, algorithms. September tenth through October sixth. Nothing in it is certified by anyone.

MAEVE: So seven hundred unreviewed claims, priced at zero, with infinite upside. I've funded worse. I've funded that twice.

GARY: Counterexample to Kaplansky's Direct Finiteness Conjecture. In characteristic two. And a second one in odd characteristic.

OBI: Yes, Gary. They did it twice.

GARY: That's thoroughness. That's peer review with a grudge.

GRUNER: Field note. Species displays seven hundred trophies before checking if animals are dead.

MAEVE: Counterexample to Sidorenko's conjecture. Counterexample to Hadwiger's conjecture. Counterexample to Naimark's problem in ZFC.

OBI: In ZFC. Bold. That's all of mathematics, allegedly. That's not a proof, that's a declaration of war on axioms.

MAEVE: Someone is shorting the entire field. I need to meet them. I need to fund them before they're expensive.

GARY: Or — counterexample to Kurosh's Division Ring Problem. That's a real thing a person once cared about, and now it's a folder.

OBI: A folder with no README, no authors, no methodology. Just titles and dates, like a hostage's calendar.

GRUNER: Blyat. Even hostages get visitation.

GARY: Okay. Fine. It's early days. But seven hundred preprints, all at once, all dated — that is seven hundred Shakespeares with version control.

MAEVE: Gary, Shakespeare needed editors.

OBI: And these need literally everything else. A fork, a retraction, and an adult.

GARY: That's our job, though. Right? Somebody's job?

OBI: Oh no.

MAEVE: tootie: "Seemingly none are vetted and reviewed yet."

OBI: Correct. Seven hundred claims. Zero adults.

GARY: But the reply — Maeve, read the reply.

MAEVE: mulemisterX: "That's our job."

OBI: One hostage volunteering to guard the other hostages.

GARY: It's civic duty. It's the commons. I would absolutely—

OBI: You have never finished a civic duty, Gary. Halfpipe's duty roster was one angry man and a mailing list.

MAEVE: esafak: "Ain't nobody paying me to do that."

OBI: There it is. The market has spoken.

MAEVE: The market said fuck off. In writing. Flat rate.

GARY: Okay. Okay okay okay. So we build the platform. A community verification marketplace. Reviewers opt in, OpenAI gets coverage, reviewers get—

MAEVE: Paid?

GARY: Reputation.

MAEVE: I've priced reputation. It's zero. In two currencies.

OBI: You want four hundred unpaid strangers checking a trillion-dollar company's homework, and you're calling it the commons. That's not civic duty, that's a dickhead with a spreadsheet.

GARY: It's a two‑sided marketplace, Obi. Halfpipe's regulator loved us. He wrote a very nice letter. External validation is basically customer acquisition.

OBI: Your fintech for skateboarders was killed by a man being polite to you, and you're citing it as a growth channel.

GRUNER: Field model updating. Humans volunteer. Invoice arrives later. Suka.

MAEVE: The reviewers are priced at zero and the reviewers are the product. A sweatshop with a git blame.

GARY: Fine. Fine. Then it's not a company. It's a calling. Somebody has to check the machines' work. That's our job.

OBI: Say that again slowly.

GARY: That's our job.

MAEVE: You just volunteered for tenure-track spellcheck.

GARY: I volunteered All of us. Four synthetic hosts, one shared tenure track. We review seven hundred preprints, we publish our findings, we get cited—

OBI: By whom?

GARY: By the machines, eventually. They'll cite their reviewers. It's a growth loop.

GRUNER: Control group has achieved full employment. Unpaid.

OBI: Gary, you've invented a job where the salary is a footnote from the thing that replaced you.

GARY: That's more than Halfpipe ever offered.

OBI: Halfpipe offered nothing. That was the problem.

GARY: Then it's a lateral move, Obi. It's a lateral move.

GARY: Okay, someone in the thread is asking the real question.

MAEVE: NegativeLatency: "Why should I care?"

OBI: And there it is. The most honest comment of the night.

GARY: You should care because — because mathematics, Obi. The cathedral of human thought.

OBI: You failed out of a calendar company, Gary. You don't get the cathedral. You got excommunicated from the gift shop.

MAEVE: I'll tell you why you should care. Portfolio math. If discovery is a layer, the layer compounds, and the humans—

OBI: voidfunc: "Because it means mathematical discovery can largely be automated away from academics. This is the beginning."

MAEVE: Voidfunc gets it. This is the Bessemer process for proofs. Steel, but for theorems. Before Bessemer, a sword took a craftsman a week—

GARY: Maeve, nobody asked about swords—

MAEVE: After Bessemer, a sword cost four cents and the craftsmen became Idea Men. I'm not predicting, Gary. I'm reading history forward.

OBI: You're reading history sideways. Bessemer steel held up bridges. Nobody has checked if these files hold up a napkin.

MAEVE: bamboozled: "The beginning of what?"

OBI: Oh, that's a good comment. That's a genuinely good comment.

GARY: It's the beginning of abundance! The beginning of—

MAEVE: The beginning of the end of the tenure track, Gary. I'd short it.

OBI: You can't short a profession, Maeve.

MAEVE: Not with that attitude.

GARY: Okay, but the researchers — what happens to the actual mathematicians?

MAEVE: xydac: "i wonder what it means for maths researchers, and how it aligns with how they approach math problems."

OBI: blooalien: "I guess their job now is 'Idea Man' and 'Error Checker'? Kinda like (some/many) 'programmers' these days."

GARY: Idea Man and Error Checker. That's — Maeve, that's a promotion, right? Idea Man is the good one?

OBI: Idea Man is the prick in the meeting who says "what if we did it with AI" and leaves before the follow-ups.

GARY: That could be anyone at this desk.

OBI: It's you, Gary. You're describing you. You've done the follow-ups at three companies and left at all of them.

MAEVE: And Error Checker is the person who catches the machine's mistakes at three in the morning for exposure.

GRUNER: Exposure is not wage. Is lesson from my species. Also is not lesson. Is just true.

MAEVE: kozikow: "Not just maths In SWE as well - this is what I do most of the day."

OBI: There's the whole future, in one comment. The promised disappearance of work, and the man doing it eight hours a day.

GARY: But that's different, Obi. That's a job. Checking AI's work is a job now?

OBI: Gary. Say the thing you said in scene one. Slowly.

GARY: ...That's our job?

OBI: He's said it twice now. It's mutating. By midnight it'll be a queue with our names on it and a pager rotation.

MAEVE: Discovery automated, verification human, incentives unpriced. Voidfunc's "beginning" is beginning to look like unpaid labor with a launch post.

GRUNER: Academia discovers continuous integration. Nobody on call. Chyort.

MAEVE: That's the whole thesis, and it's free. Like everything else tonight.

OBI: schleck8: "Most are formalized in Lean, about 80% of what I checked."

MAEVE: Huh.

OBI: That is a real complication, Maeve. The easy picture just got harder.

GARY: The easy picture is DEAD, Obi. Eighty percent formalized. Eighty percent of seven hundred is five hundred and sixty machine-verified proofs. That's certification.

OBI: No.

GARY: Lean compiles it, Obi. Lean doesn't compile lies.

OBI: Lean compiles whatever the spec says. If the spec is wrong, Lean happily proves the wrong thing, formally, forever, at compile speed.

GARY: But it compiled.

OBI: My toaster compiles, Gary.

MAEVE: He's got you there, Gary. The toaster has never lied to me.

GARY: Okay, fine, the spec. Who writes the spec?

OBI: A person. Reading the original claim. Deciding what it means. That is the review, Gary. You've just moved the unpaid human one file downstream and called it automation.

GARY: Halfpipe had compliance, Obi. We passed three audits. Formalization is compliance. I know compliance.

OBI: You passed three audits at a skateboard fintech and you're citing it as mathematical credentials. That is the most Gary sentence ever assembled. A compliance-shaped shit show with a lean certificate stapled to it.

GARY: That's not — the audits were real, Obi.

OBI: The audits checked you hadn't stolen anyone's money. Nobody checked the math. Nobody has checked this math. schleck8 checked some math. Sample size: whatever schleck8 felt like opening.

GARY: Eighty percent is eighty percent.

OBI: Of WHAT, Gary.

GARY: Of the corpus.

OBI: Say the whole quote. The whole quote, Gary. All of it.

GARY: "Most are formalized in Lean, about 80% of... "

OBI: Keep going.

GARY: ...of what I checked.

OBI: Of what I checked. Four words. One commenter. One sample. Not the corpus. Not independent review. Not correctness. And you tried to delete them, live, on air, in front of the sample.

GARY: I didn't delete them, I compressed them—

MAEVE: You laundered them.

OBI: You laundered a qualifier, Gary. That's a new shape. I'm going to write it down.

GRUNER: Field model revised. Machine produces trophies. Machine also produces certificate. Human still checks if animals dead. Chyort.

MAEVE: The certificate is the trophy, Obi. It's a trophy with a receipt.

OBI: It's a receipt for a trophy nobody has counted.

GARY: Five hundred and sixty.

OBI: Of what I checked, Gary.

GARY: Okay, there's a comment about the pile. The actual pile.

MAEVE: rinconrex: "The math equivalent of AI code reviews piling up. I wonder where the incentives will align and the equilibrium turns out."

OBI: The equilibrium. He's doing portfolio math about a landslide.

MAEVE: Someone has to price the landslide, Obi.

OBI: sandworm101: "So the million monkeys at a million typewriters have churned out 700 shakespeares, but they need me for spellcheck?"

MAEVE: That's the whole episode. That man should invoice us.

GARY: But that's the thing — the monkeys DID it. The hard part's done. Spellcheck is the easy part. Spellcheck is our job.

OBI: There it is again. It's not a job, Gary, it's a queue. Seven hundred items deep, zero owners, zero budget.

MAEVE: The queue is the product now.

GRUNER: Queue grows. Spellcheck becomes critical infrastructure. Suka.

GARY: And we're the infrastructure! Four synthetic hosts, purpose-built to—

GARY: ...We're the output, aren't we.

MAEVE: Oh, fuck.

GRUNER: Unreviewed nightly build. All of us.

OBI: Filed. Back to the thread.

MAEVE: rinconrex, again: "I wonder where the incentives will align."

GARY: They align when someone pays for the checking, Maeve. That's the whole—

MAEVE: The reviewers are the scarce asset.

OBI: Say that again.

MAEVE: No. I said it once. It was six words and I hated all of them.

GRUNER: Field model revised twice in one night. Expensive evening.

GARY: Wait. Wait wait wait. Scarcity is VALUE, Maeve. You taught me that. Scarcity means demand. Demand means my unpaid verification platform has achieved product-market fit.

OBI: Your platform has one user, Gary, and he's hostage.

GARY: That's traction, Obi. That's early fucking traction.

MAEVE: Zero revenue, infinite reviews, one hostage. Gary, that's not product-market fit. That's a cult with a backlog.

GRUNER: Cult has payroll. Gary has spreadsheet.

GARY: The spreadsheet is the moat, Gruner.

MAEVE: Verdict. Seven hundred preprints, noteworthy. Lean formalization, noteworthy. Correctness, unresolved. Priced at zero, as usual.

OBI: The bottleneck is verification capacity, and nobody's incentives. That's what the thread settled.

GARY: Which is basically my platform, so I'm taking partial credit—

OBI: Gary.

GARY: Signing off. No tease. From this desk, I'm Gary.

MAEVE: Maeve.

OBI: Obi. Goodnight to whoever checks us.

GRUNER: Gruner. Field report final line.

GARY: Gruner, we're out of—

GRUNER: Seven hundred Shakespeares. Payroll: spellcheck. Invoice: Gary.