AI Keeps Pissing Off Mathematicians

Summary
Presented by Dagogo Altraide on ColdFusion, this video explores the tensions between frontier AI labs and the academic mathematics community, focusing on OpenAI’s claimed breakthrough on the Navier–Stokes Millennium Prize problem. Altraide details allegations by NYU mathematician Tristan Buckmaster that OpenAI may have derived its solution from his unpublished research sessions in Codex, alongside broader concerns regarding how AI labs leverage researcher data.

What is shown

• OpenAI’s October 6, 2026 GitHub release (“Sharing AI progress in mathematics”) cataloging 722 manuscripts across 372 result families [00:17].
• Video reactions from mathematicians, including Dr. Samuel Allen Alexander celebrating results and others critiquing frontier labs’ approach to research [00:30, 00:45].
• Visualizations of fluid dynamics, vortex stretching, and singularity blow-up modeled by the Navier–Stokes equations [04:18, 05:44].
• Clips from an interview between Brady Haran of Numberphile and mathematician Tristan Buckmaster discussing his collaboration with Levent Alpöge and interactions with OpenAI [07:35, 08:39, 13:21].
• OpenAI’s technical report and execution metrics for its multi-agent math solver run [10:13].
• Excerpts from Buckmaster’s public statement, news articles from The New York Times, and commentary from Terence Tao on Hacker News and Mastodon [11:13, 14:13, 16:22].
• A sponsored demonstration of the interactive STEM learning platform Brilliant featuring the AI tutor “Koji” [20:54].

Claims & numbers

• OpenAI claimed to solve the Navier–Stokes existence and smoothness Millennium Prize problem on September 8, 2026, using an internal model more capable than GPT-6 Astra [02:43].
• The Clay Mathematics Institute established a $1,000,000 bounty on each of the seven Millennium Prize Problems in 2000; only the Poincaré conjecture had previously been solved (by Grigori Perelman) [04:47].
• Buckmaster and Alpöge used Anthropic’s Claude and OpenAI’s Codex (specifically GPT-5.6 Sol and Astra) to construct a forced blow-up candidate between mid-August and late August 2026 [07:18, 08:00].
• OpenAI deployed an internal multi-agent system comprising roughly 10,000 concurrent agents running for about 88 hours, producing 2.7 million messages and 130 billion output tokens, followed by 17 hours of formalization in Lean by GPT-6 Astra to produce a ~100-page proof [10:14, 10:31].
• OpenAI stated that no user inputs past a July 3rd cutoff influenced the system, denying unauthorized use of Buckmaster’s Codex chat logs [15:30].
• Terence Tao likened AI labs rushing for prize proofs to “using excavators to loot an archaeological site” [16:47].

Notable quotes

• “The first LLM generated proof Levent sent me was the most horrendous I have ever read.” — Tristan Buckmaster [08:00]
• “Why would you ruin your career?” — Alleged message from OpenAI to Tristan Buckmaster [13:04]
• “Using excavators to loot an archaeological site.” — Terence Tao [16:47]

Assessment
This is an analytical journalistic overview combining interview footage, public documentation, and news reporting to evaluate the controversy between academic researchers and OpenAI. The video relies on reported claims and interviews rather than presenting an independent technical verification of either party’s mathematical proofs.

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