OpenAI’s new massive math drop:
OpenAI@OpenAI
We publish a wide range of new mathematical results produced by an internal boundary model. We consulted with the Institute for Advanced Study’s Independent Mathematics and Artificial Intelligence Advisory Group, and we leveraged their advice and…
10:19 p.m. · October 6, 2026 · 15.2 million Views
1.41K Answers · 5.3K Reposts · 33.3K Tastes
(This essay was written in extreme haste before a very long flight without Wi-Fi; please forgive the typos.)
First part: my opinion
The real news here is not the outcome; that’s not what we were told.
1. AI once tried to be a science. Now we’re getting things like the completely vague report from OpenAI below:
“Same procedure”? “Use a new model”?
This would never pass peer review.
We don’t know what the procedure was.
We don’t know anything about architecture. For example, were the proofs generated all at once and then verified by the Lean token system? Was there an iterative process?)
We don’t know anything about the failure rate. We don’t know anything about training/post-training/data augmentation.
2. As a result, we have no idea how generalizable the result is outside of mathematics.
3. Much of the discussion on social media has been reduced to a section of ignorant cheerers clapping without knowing what it means or what it could mean – without ever asking fundamental scientific questions.
The new system could be a legitimate step toward AGI. Or it could simply be intelligent exploitation of Lean and synthetic data in a verifiable domain without any generality.
From the initial report we can say almost nothing.
§
Part two: Terence Tao’s point of view
Gn bussni
