OpenAI Just Broke Math With Its Most Powerful AI Yet

Summary
This video, produced by the channel AI Revolution with voiceover narration and motion graphics, reports on OpenAI’s October 6, 2026 release of 722 AI-written mathematics papers to GitHub. It breaks down the claimed breakthroughs—ranging from the Quasi-Riemann hypothesis to the Unique Games Conjecture—and examines the intense friction, priority disputes, and ethical pushback emerging from the academic mathematics community.

What is shown

• [00:10] The GitHub repository openai/math displaying the manuscript collection index of 722 papers across 372 result families.
• [00:49] The Lean interactive theorem prover website and code interface verifying mathematical proofs.
• [00:55] Title and abstract of the preprint “The Quasi-Riemann Hypothesis: A Zero-Free Half-Plane $\Re s > 7/8$” dated September 30, 2026.
• [01:29] Clay Mathematics Institute pages covering Millennium Prize problems, including the Navier–Stokes equation and the Hodge conjecture.
• [01:39] Interface clips of ChatGPT Work and Codex completing tasks with multi-hour thinking runs.
• [01:53] Sam Altman’s post on X announcing “Sharing AI progress in mathematics”.
• [03:18] Title and abstract of “The Unique Games Theorem” manuscript dated September 23, 2026.
• [04:23] Paper excerpt claiming proof that Catalan’s constant is irrational.
• [04:49] Reporting from The Washington Post, Wired, and The New York Times documenting tensions between mathematicians and OpenAI researchers.
• [06:54] Excerpts from open statements, including “A Severe Misalignment of AI in Mathematics” and recommendations from the Advisory Group on Mathematics and Artificial Intelligence.
• [09:52] Christian Szegedy’s essay on X analyzing the future trajectory of human mathematicians versus AI systems.

Claims & numbers

• The narrator states that on October 6, OpenAI uploaded 722 mathematical manuscripts grouped into 372 result families to GitHub (openai/math) [00:10, 00:35].
• The narrator states that 235 of those families include formal, computer-checked proofs in the Lean programming language [00:46].
• The narrator says OpenAI evaluated approximately 4,000 real unsolved math problems, publishing only the significant results [01:10].
• The narrator states the model began training on August 28, solved a Navier–Stokes problem on September 5 using 10,000 AI agents running for 88 hours, but for this release nearly every result was produced by a single agent with a single prompt averaging ~3 hours of ChatGPT Pro thinking [01:23, 01:35, 01:43].
• The narrator states the model proved a zero-free half-plane $\Re(s) > 7/8$ for Dirichlet L-functions and uniformly excluded Landau–Siegel zeros [00:55, 02:24].
• The narrator mentions that roughly 40 mathematicians were convened in a closed-door summit by OpenAI in August [04:49].
• The narrator reports that the Leiden Declaration on AI in mathematics garnered signatures from over 4,000 mathematicians [09:01].

Notable quotes

• [05:13] Bryna Kra: “The room felt like extreme excitement and extreme fear at the same time.”
• [06:27] Sébastien Bubeck (reported quote): “If you don’t want me to be polite, I don’t have to be polite.”
• [08:37] Andrew Sutherland: “We should demand to see the receipts.”

Assessment
This video is a journalistic summary and commentary aggregating public preprints, news reporting, and community statements. It does not contain an independent hands-on evaluation of the underlying proprietary frontier model, relying instead on OpenAI’s repository disclosures, academic preprints, and press coverage.

Described by gemini-3.8-flash on 2026-10-08 from the video’s audio and frames.