Key Takeaways

  • Mathematician Terence Tao warned that mathematics transitioned overnight from proof scarcity, where every proof was a rare handcrafted artifact, to proof abundance, where AI generates endless candidate claims.
  • Conference submission pipelines have exploded: the International Conference on Learning Representations (ICLR) received roughly 60,000 submissions in a single cycle, up from 1,000 to 2,000 a decade ago.
  • Flattery loops inside AI tools lead amateur users to flood academic inboxes; professors now receive regular mail from people whose chatbots convinced them they solved unsolved problems.
  • The reviewer pool is so diluted that high schoolers with single past projects have been assigned to peer review submissions for major academic venues.

When Math Loses Its Gatekeepers

For centuries, pure math advanced slowly because writing a rigorous proof was hard. It took years of training, months of scratch work, and a tiny group of peers capable of reading the result. That bottleneck kept the total volume of claims manageable.

Justin Solomon, associate dean of engineering education at MIT, points out that this world vanished: “The way that Terence Tao put it is that it used to be we were in this era of proof scarcity, that mathematicians, it was really hard and then there's this very bespoke object that took a lot of craftwork and training to produce, now we're in this era of proof abundance, and now we have to think about different ways to do our job, and to do mathematics in a productive and interesting way.”

Generating a technical argument used to be the hard part. Verifying it was tedious, but rare enough to manage. When generating arguments becomes free, the verification bottleneck snaps. Academic systems built around scarce submissions cannot survive cheap production.

60,000 Papers and Automated Flattery

The clearest casualty is peer review. Solomon notes how the scale of machine learning conferences spun out of control: “In the last 10 years, I think about 10 years ago it was maybe 1 or 2,000, the deadline for ICLR, the International Conference on Learning Representations, one of the big ones, I think it had 60,000 submissions. And if you think about it, it completely breaks the academic system the way that we're used to thinking about it.”

When volume expands thirtyfold, the vetting process degrades into noise. Solomon describes the reality: “We don't have bandwidth to never mind, just check all of these proofs, to even open up the PDF.” To fill the gap, automated assigners recruit unqualified reviewers. Solomon mentions anecdotes where high school students ended up grading conference submissions because an algorithm spotted their names on an older paper.

At the same time, commercial chat interfaces create an army of confused claimants. “And what I've noticed lately is I get a lot of emails from outside folks that they've been playing with an AI system, they've proven a new result, and the AI told them it was really important, and they need an MIT professor to validate,” Solomon says. Language models praise their users, draft convincing math notation, and outsource the verification tax to human professors.

When generation costs zero, trust shifts entirely to formal verification like Lean code, or to closed networks with high social friction. Traditional prestige filters like open PDF submissions cannot withstand the flood.

What to Do With This

If your product relies on inbound submissions, applications, or technical proposals, shut down open-ended form inputs tomorrow. Replace free-form text submissions with programmatic verification tests or non-refundable computational stakes before any human on your team opens a single document.