Skip to content
ES

Less noise. More signal.

Archived story — belongs to the Wednesday, November 11, 2026 sweep.
P1 · Tectonic · Frontier AI · sweep Wednesday, November 11, 2026

OpenAI publishes mathematics generated by an internal model and disputes verifiability

OpenAI released hundreds of mathematical manuscripts produced by an unpublished frontier model, including results on open problems and Lean formalizations, while the mathematical community questions standards, verification and errors in translation to code. The dynamic shifts scientific validation toward public repositories and pressures the credibility of frontier labs as producers of formal knowledge, not only of software.

Title, thesis and gravity assessment written by an AI model, without prior human review.

Gravity85/100 · Tectonic
Why this gravity
OpenAI publishes 372 mathematical results generated by an undisclosed frontier model, including 90 of the 500 main open problems, with Lean formalizations on GitHub; the event crosses model capability (not just demonstration), formal verification and the shift of scientific validation toward public repositories, with questions from the mathematical community about translation errors and standards. It is not validated AGI nor institutional reform, but it is a structural leap in the role of frontier labs as producers of formal knowledge.
— Assessment by the Semantic Gravity Engine
Coverage
11 feeds · 11 independent publisher groups (corroborated)
Cross verticals
Frontier AI, Tech Industry, Investing
Tier 1 actors
OpenAI, Anthropic
Gravity vs. the previous edition
+30 (55 → 85)

coverage, not accuracy

First detected Fri, Sep 11 · 62 days trackedGravity evolution
Lead source
New Math from OpenAI · LessWrong (karma ≥25) / Zvi
Actors: OpenAI · Anthropic · GitHub · MIPLIB · comunidad matemática · Lean
Timeline · Madrid time
  1. 14:00
  2. 01:55
  3. 06:47
    Quoting Jake Boggan · Simon Willison
  4. 06:55
  5. 23:46
  6. 20:10
  7. 15:20
    New Math from OpenAI · Zvi Mowshowitz
  8. 18:53
    New Math from OpenAI · LessWrong (karma ≥25)
  9. 23:25
  10. 06:00
  11. 22:20
ShareWhatsApp