Skip to content

Welcome to the AI crisis in math

Decoder

40m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

The episode explores the sudden existential crisis facing academic mathematicians following OpenAI's publication of ten advanced math and theoretical computer science proofs generated by a frontier model named Astra. Host Nilay Patel and guest Robert Hart discuss how these models have jumped from terrible performance in basic arithmetic to successfully tackling sophisticated, decades-old problems like the unit distance conjecture. While some leading mathematicians view the breakthroughs with professional awe, others worry about hype, inadequate attribution, and whether automating away problem-solving will sterilize the field or leave graduate students without a viable career path.

The discussion unpacks the tension between the pure logic required for mathematical proofs and the opaque methods used by AI labs to achieve them. Hart explains that while mathematicians appreciate the rigor verified through tools like Lean, the speed and secrecy surrounding unreleased models have left the academic community in a state of shell shock. Ultimately, the episode highlights the uneasy collision between an ancient academic discipline and corporate AI labs using mathematics as an advertising playground.

  • Frontier models have rapidly advanced from struggling with basic arithmetic to successfully tackling high-end abstract mathematics at a professional level. 3:48
  • OpenAI recently published documentation claiming an internal model named Astra solved ten major advances across various theoretical disciplines. 9:12
  • Mathematicians use a formal programming language and computational prover called Lean to test the rigor and assumptions of automated proofs. 17:08
  • Some academic researchers have signed the Leyden Declaration, an open letter pledging not to buy into industry hype surrounding artificial intelligence. 33:07

Chapters

  1. 0:01
    Intro and Host Greeting

    The host introduces the episode and the topic of AI’s impact on mathematics.

  2. 1:57
    AI’s New Math Capabilities Announced

    The host explains OpenAI’s recent breakthroughs that have sparked debate in the math community.

  3. 2:57
    Interview with Robert Hart Begins

    The host begins the conversation with The Verge’s AI reporter, Robert Hart, about AI’s effect on math.

  4. 3:44
    Mathematicians’ Existential Crisis Discussed

    Hart summarizes mathematicians’ concerns about AI solving advanced problems and the potential loss of human contribution.

  5. 8:40
    Astra’s 10 Breakthroughs Highlighted

    The hosts discuss OpenAI’s Astra model and the ten significant mathematical achievements it claims to have solved.

  6. 12:24
    Credibility and Attribution Concerns

    Mathematicians express mixed feelings about OpenAI’s crediting practices and the transparency of its proofs.

  7. 27:17
    Impact on Academic Careers and Funding

    The discussion turns to how AI may affect research funding, career trajectories, and the value of human mathematicians.

  8. 30:09
    Democratization of Math vs. Quality Control

    The hosts debate whether AI tools broaden access to mathematics or flood the field with unverified results.

  9. 39:53
    Future of Mathematical Research and Innovation

    The episode concludes with reflections on whether AI will stifle or stimulate new areas of mathematical inquiry and the broader implications for the discipline.

Highlights

  • 3:45

    Existential Crisis in Mathematicsinsight

    The guest explains that a recent phase transition in AI capabilities has sparked an existential crisis among mathematicians regarding their future role.

  • 4:54

    Terrible at Arithmetic, Good at Mathdata

    Models remain notoriously poor at basic counting, days of the week, and simple arithmetic while performing impressively on abstract academic papers.

  • 8:46

    Disproving an 80-Year-Old Problemdata

    An internal OpenAI model successfully disproved the unit distance conjecture, an eighty-year-old mathematical problem.

  • 9:12

    OpenAI Drops Ten Major Proofsstory

    OpenAI released hundreds of pages detailing ten advances solved by their internal model Astra in fields like quantum game theory and sphere packing.

  • 13:08

    Sloppy Attribution in AI Papersdebate

    OpenAI initially claimed no progress had been made on these problems in a decade, overlooking prior work by human researchers until quiet corrections were made.

  • 20:59

    Mowing Down Math Problemsinsight

    A researcher warns that AI might mow down existing problems without actually pushing the field forward if humans are removed from the loop.

  • 24:07

    Shell Shocked Academic Communityinsight

    The overriding feeling among academic mathematicians interviewed for the story is one of shell shock due to how rapidly the landscape has shifted.

Quotes

  • “If AI can do math of this caliber, does that mean AI labs can transfer those skills to other domains?”
    Unnamed speaker · 2:21
  • “They are still truly, truly terrible at some areas of math.”
    Unnamed speaker · 4:54
  • “I think it's a lot of what these other fields have been struggling to deal with for the last five years in a very compressed period of time.”
    Unnamed speaker · 4:18
  • AMEX Corporate 0:01–0:35
  • Midi 0:42–1:05
  • Michelob Ultra 11:21–11:44
  • Hilton 26:04–26:35
  • Indeed 26:40–27:13
  • Milo's 41:40–42:07

About this episode

My guest today is Robert Hart, The Verge’s London-based AI reporter. Robert recently wrote a fantastic story for us about the debate raging inside the world of mathematics — and the existential crisis over what it means that new frontier AI models have become very good at math in a shockingly short period of time. I wanted to dive into all of this with Robert, who actually spoke to some of the most accomplished mathematicians working today to figure out what’s hype and what’s real — and to get a sense of just how exciting, and how scary, all of this is. Read the full interview transcript on The Verge. Links: The AI takeover of mathematics has begun | The Verge Ten advances in mathematics and theoretical computer science | OpenAI An unreleased Anthropic model made progress on the Riemann hypothesis| TechCrunch OpenAI’s amazing — but vastly oversold — new model Astra | Gary Marcus OpenAI’s math breakthrough played to AI’s strengths | Understanding AI Why the legendary Erdős problems are falling to AI | Quanta Subscribe to The Verge to access the ad-free version of Decoder! Credits: Decoder is a production of The Verge and part of the Vox Media Podcast Network. Decoder is produced by Kate Cox and Nick Statt. This episode was edited by Ursa Wright. Our supervising producer is Greg Ott, and our editorial director is Kevin McShane. The Decoder music is by Breakmaster Cylinder. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Transcript

0:00
40:30

8,314 words · 9 speakers

Topics

  • Decoder
  • OpenAI
  • Astra
  • The Verge
  • Apple Watch Series 12

Page updated · transcript and summary are machine-generated