What OpenAI announced
In an October 26 announcement, OpenAI says its internal models produced results concerning local smoothing in every dimension, a refutation of the Exponential Time Hypothesis (an assumption about how fast certain satisfiability problems can be solved), and rigorous existence results for quantum chromodynamics and related field theories. These are OpenAI’s research claims, with academic groups helping examine and explain the work. [1]
The company describes the manuscripts as raw research data. It has published natural-language and Lean materials for the mathematics and computer-science results; the physics group is holding back its AI-produced artifacts while preparing its own write-up. The announcement gives a date, but we did not establish its posting hour. This is a catch-up briefing. [1]
Read the statement that was formalized
Lean is a proof-assistant system. The linked documentation describes the scope of each formalization. The local-smoothing document specifies mathematical estimates and comparison statements. The ETH document distinguishes the formalized running-time bound from a sharper bound in the accompanying manuscript. Those distinctions matter when interpreting a headline about a proof. [2][3]
We read those scope documents, but did not compile the proofs or independently validate the mathematics. This announcement does not establish a resolution of P versus NP, an immediately faster business application, or access to the internal research models. [1][3]
Our take
For AI enthusiasts, the useful lesson is the chain of evidence: who makes the claim, what artifact is available, what exact statement it covers, and what independent experts can explain or reproduce. A formal proof is valuable evidence about its formal statement; understanding how that statement matches the intended scientific question is also essential.
For a small business, keep watching without changing operations on the strength of the headline. Progress on difficult research problems is worth following, but it does not measure whether an everyday assistant will handle your invoice, spreadsheet, or customer message correctly. Check the result of the actual task you need done.