OpenAI dropped 722 math papers. Mathematicians are split.

Summary
This video is an excerpt from the ThursdAI podcast hosted by Alex Volkov alongside panel members Peter, Yam, and LDJ. The panel examines OpenAI’s GitHub release of 722 research-level mathematics manuscripts produced by an unreleased frontier reasoning model, detailing the specific conjectures addressed, the computational effort required, missing cryptography results, and the polarized reactions within the mathematical community.

---

What is shown

• [00:40 - 00:47] ThursdAI title sequence displaying model releases and benchmark charts (“Number Bench”).
• [05:08 - 05:34] Alex shares a screenshot of an elementary grade-school word problem about Cindy’s textbook weights to contrast past LLM capabilities with current performance.
• [11:21 - 11:58] Alex shares a post on X by mathematician Kevin Buzzard titled “Understanding mathematics,” discussing whether automated proof generation equals mathematical understanding.
• [14:18 - 14:37] LDJ presents an X post and an annotated list compiling the top 100 open mathematical problems, highlighting which ones were claimed solved by OpenAI.
• [14:57 - 15:32] LDJ navigates OpenAI’s openai/math GitHub repository and the overview.pdf document (“OpenAI Research Catalog”), showing numbered results and pointing out missing indices (such as #45) and the notable absence of cryptography-related results.

---

Claims & numbers

• OpenAI Math Manuscripts Release:
  • The hosts state OpenAI dropped 722 math manuscripts across 372 result families on GitHub on Tuesday night without significant fanfare.
  • Alex notes the internal frontier model tackled approximately 4,000 open research problems, averaging 3 hours of compute per result at a ChatGPT Pro reasoning tier.
  • Specific problems addressed include the quasi-Riemann hypothesis (zero-free region with human-edited write-ups), the Hodge conjecture for CM abelian varieties, irrationality exponent of $\pi$, Mahler conjectures, Kaplansky conjecture, and Khot’s Unique Games Conjecture.
  • LDJ states that out of an open problem set of 500 problems compiled using Fable and Astra, 92 were resolved in the release (including 10 of the top 100).
  • LDJ notes that ~20% of the resolved results consisted of counterexamples rather than constructive proofs.
• Missing Results & Cryptography:
  • LDJ notes several sequence gaps in the numbering of OpenAI’s GitHub catalog (e.g., skips between 44 and 46), claiming that cryptography-related mathematical results were omitted, likely due to national security sensitivities or export policies.
• Mathematician Reactions:
  • Peter references a letter from the Association for Human Mathematicians (reposted by Terence Tao) representing ~800 members, which criticized the release and rejected the claim that autonomous problem-solving necessarily advances mathematical understanding.

---

Notable quotes

• [01:06] Alex Volkov: “5 moments in one drop. If you guys remember the Navier-Stokes and how much hype that did, we just got 5x in the amount of OpenAI solved maths.”
• [08:21] Yam: “They went after the Riemann hypothesis! That’s insane on its own, just to try.”
• [18:06] Peter: “We reject OpenAI’s assertion that this release advances our subject... which goes back to the question of what is maths for?”

---

Assessment
This is a podcast review and news commentary episode discussing a verified technical drop from OpenAI. The hosts display real external websites, GitHub repositories, and social posts to contextualize the release, offering analytical opinions without promotional staging.

Described by gemini-3.8-flash on 2026-10-15 from the video’s audio and frames.