Drowning in a Sea of Proofs

Overview

The video discusses two recent developments in AI-assisted mathematics, both set against a backdrop of growing backlash from parts of the mathematical community. The backlash centers on concerns about how AI systems publish and claim results.

---

1) OpenAI claims major progress in solving mathematics

• OpenAI released an update dated 21 September 2026, titled “Advisory Group on Mathematics and Artificial Intelligence.”
• The update claims that an internal model—trained starting 28 August—has:
  • Resolved the Navier–Stokes Millennium Prize problem
  • Solved “more than 100 long-standing open problems” across many areas of mathematics

How the creator interprets the claim

• The speaker argues that the claim would be extraordinary.
• They suggest it is likely true in part because a false claim could trigger significant PR backlash, especially given existing criticism from mathematicians.
• The video also reports that OpenAI acknowledges the speed and character of the progress surprised mathematicians inside OpenAI, prompting internal discussions about how to communicate and adapt responsibly.
• The announcement frames AI’s mathematical capabilities as broadly scientifically important, and positions OpenAI as seeking responsible development and deployment.

---

2) Response to criticism about “negative externalities” of AI benchmarks

• The video references an earlier widely discussed statement/letter: “A severe misalignment of AI in mathematics” by Terrence Tao and other prominent mathematicians.
• The creator argues OpenAI appears to be taking these concerns seriously, particularly:
  • The harm that can come from using already-solved problems as benchmarks
  • Damage to disciplinary norms, including how results are validated, attributed, and disseminated
• The video characterizes OpenAI’s response as shifting toward engagement rather than exclusion—working to involve mathematicians in how results are reviewed and communicated.

---

3) Creation of an independent mathematics advisory group

OpenAI announces an independent mathematics advisory group intended to function as a bridge between AI labs and the broader mathematics community.

Stated responsibilities

The group is said to advise on:

• Reviewing and communicating emerging AI results
• Assessing the significance of new theorems
• Coordinating dissemination
• Academic/professional standards for mathematical research

How the group is described

• Independent from OpenAI
• Unpaid by OpenAI
• Able to make advice public, including unsolicited feedback
• Not responsible for advising OpenAI on how to pace its internal progress

---

4) Guest blog: why a group member joined

The creator reads from a guest blog post by mathematician Martin H. (spelled in subtitles as “Martin Herer”): “Why I agreed to join AGM AI.”

Main claims from the blog

• The group is genuinely independent of OpenAI and other frontier AI labs.
• Not all members were necessarily contacted first by OpenAI.
• The nine mathematicians are the only ones involved in their direct conversations.
• Technical support (so far) is mostly administrative/logistical via an organization labeled IAS (abbreviation explained on-screen in the video).
• Members did not sign special constraints limiting what they can say publicly (aside from normal confidentiality expectations).

Reasons given for joining despite risks

1. AI-related math results—especially on private websites—have sometimes been associated with poor scholarship and less effort than typical standards.
2. AI has already substantially impacted math research, raising issues of attribution, priority, and understanding, making refusal feel unrealistic.
3. Avoiding contact could allow AI companies to depict mathematicians as out-of-touch—an outcome the author wants to avoid.
4. Since the “misalignment” statement calls for urgent community/company action, joining an advisory channel is framed as a way to help address the problems rather than protest from the sidelines.

Uncertainty and next steps

• The blog acknowledges uncertainty: the group doesn’t yet know the best way to handle the unprecedented situation.
• It plans to collect feedback via forms and discussions before finalizing guidance.

Criticism and the creator’s framing

• The video maker describes commenters who strongly disagree, arguing AI labs should not be legitimized.
• The creator frames the advisory group as potentially “a step in the right direction,” while emphasizing the need to see what concrete actions it will take.

---

Key takeaway

The video presents OpenAI’s announcement and the creation of a mathematics advisory group as an attempt to address mathematicians’ complaints about AI systems solving and publishing results faster than normal academic standards allow.

At the same time, it stresses that the practical impact remains unclear—especially regarding peer review, dissemination norms, and changes to scholarly practices.

---

Presenters or contributors

• Terrence Tao (mentioned via the “severe misalignment” essay)
• OpenAI (organization; author of the advisory-group announcement)
• Martin H. Herer / Martin H. (guest blog author; subtitles show “Martin Herer”)
• François Charles
• Camilo de Léllis (subtitles appear as “Camilo de Leis”)
• Timothy Gowers (subtitles appear as “Timothy Gabbers” / “Timothy Gvers”)
• Martin Her Nikil Shreava** (subtitles appear corrupted; likely refers to a different mathematician)
• Olri Tman (subtitles corrupted)
• Ravi Vakiel
• Edward Witten
• Melanie Matchetwood (subtitles corrupted)