OpenAI opens its Navier–Stokes work to inspection
– · 2 min read
A retrospective selection, researched on .
A mathematics publication and a new data agent may look far apart. In both cases, the important step lies between a persuasive statement and evidence you can inspect.
Mathematics: the exact problem belongs to the result
On 8 September, OpenAI shares its Navier–Stokes paper and formal artifacts. The equations describe fluid motion. The specific claim concerns smooth external forcing; it must not be rewritten as the different unforced problem. Original publication. Our article about AI and proofs explains the conditions and the difference between a test, a counterexample and a general proof.
Data analysis: an answer needs the right data
On 10 September, OpenAI introduces the Data agent in ChatGPT Work, connecting approved company data and context with analyses and interactive outputs. Existing data permissions and business definitions are part of the setup. Product announcement.
Our example: “Why did revenue fall?” The assignment needs revenue data, comparison periods and a definition of gross or net revenue. A tool can analyze the data and produce a chart and explanation. A decline alone does not establish its cause: accounting might have changed, a customer might be missing, or a season might have ended. Those alternatives need checking.
A persuasive statement starts the review
For the mathematical result, we ask about the claim, assumptions and proof. For the analysis, we ask about the data, definitions and calculation. They require different checks; the shared habit is concrete: open the artifact behind the conclusion.
A specialist can assess a proof. A sales team can inspect sample records and the chosen period. Both need more than “the AI solved it.” This week shows why a reviewable work product is itself an important part of an AI result.
Open the evidence behind a major result: the precise statement for a proof, and the data and calculation for an analysis.