AI Can Write Proofs Now. The Hard Part Is Trusting Them

Описание: On October 6th, 2026, OpenAI uploaded 722 math papers written by an unreleased AI model — in one day. Some claim huge results. None of them had been peer reviewed. So who checks them?

This video is about the hidden variable behind the headlines: verification. Why checking a proof is slow, how one minus sign took down three papers, what the proof-checker Lean can (and can't) do, and what this means for the future of mathematics.

Chapters
0:00 The drop: 722 papers in one day
0:56 Solving is cheap. Checking isn't.
1:59 One minus sign
2:53 The proof-checking machine (Lean)
4:02 What changes now

Sources
OpenAI's repository (openai/math): https://github.com/openai/math
AI Weekly (citing Nature): https://aiweekly.co/alerts/openai-pos...
ThePrint: https://theprint.in/tech/weeks-after-...
Let's Data Science: https://letsdatascience.com/blog/open...
Mezha: https://mezha.ua/en/news/openai-mathe...
Mezha (mathematicians' letter): https://mezha.ua/en/news/mathematicia...
GadgetReview: https://www.gadgetreview.com/?p=417858
IT之家: https://ithome.com/1/010/137.htm

The "two weeks per paper" figure is an illustration, not a statistic.
Animation made with Manim. Narration voice generated with ElevenLabs.

#math #AI #mathematics