A
Never breaks
For fluid filling ordinary space.
No outside push
01 / ELI5 LAB FIELD NOTE
ACTIVE MECHANISM
01QUESTION
02MECHANISM
03INTUITION
Navier–Stokes, without the fog
What OpenAI says it solved
It did not prove that ordinary water suddenly goes infinite. It says it built one carefully pushed vortex that makes the fluid equations break.
The key twist: that outside push makes the result less natural—but the official prize wording explicitly allows it.
OpenAI’s pictured flow
01 · The problem
Can the fluid rulebook break?
The equations predict how liquid or gas moves. The open question is whether a smooth flow always stays mathematically sensible—or can demand infinite speed after a finite time.
02 · Three ways to hear it
The headline hides the exact claim
Switch lenses. The argument changes once “solved Navier–Stokes” is replaced with the actual theorem and the actual prize wording.
What people hear
“AI solved how fluids behave.”
That sounds broad: normal fluid, no special help, mystery finished. It is not what OpenAI’s theorem says.
03 · The rulebook
The prize has four doors
Clay’s official statement says a solver may walk through any one. OpenAI says it used the two blue doors.
B
Never breaks
For a repeating, box-like space.
No outside push
C
Find one break
For fluid filling ordinary space.
Smooth push allowed
D
Find one break
For a repeating, box-like space.
Smooth push allowed
Important: Mehdi calls the force a loophole. It may be a less natural route, but it is written into the official problem.
04 · OpenAI’s construction
One vortex, squeezed to failure
The paper builds a very specific flow. Each step is arranged so the centre gets faster while the violent region gets smaller.
1 / 5
Starts still
The chosen fluid begins at rest.
2 / 5
Gets a smooth push
Finite, carefully shaped, limited in place and time.
3 / 5
Spirals inward
The rotating core tightens like a skater pulling in.
4 / 5
Stretches long
The same core is pulled along its axis.
5 / 5
Speed has no ceiling
At one future time, the predicted peak goes unlimited.
05 · The strange balance
Wild speed, bounded total motion
This sounds impossible until you notice that the extreme motion is squeezed into a spot racing toward zero size.
faster
The peak shoots upward
Inside the centre, the predicted speed keeps growing.
×
tinier
The hot spot collapses
The region carrying that extreme speed shrinks even faster.
The centre fails · the total energy can stay finite
06 · The technical heart
The push cannot cheat
It is easy to imagine an exploding flow if the outside push also explodes. Clay does not allow that. OpenAI says its force stays smooth.
Not allowed
Hide the explosion inside the push
That would merely command the answer and would fail the smooth-force condition.
OpenAI’s claim
Let tiny internal waves do the cancelling
The paper adds small, fast flow patterns whose momentum cancels the bad leftovers around the main vortex.
07 · Scope
What the claim says—and doesn’t
The cleanest way to avoid hype is to keep the narrow theorem beside the conclusions it cannot support.
OpenAI does claim
✓One fluid starts smoothly—from rest.
✓A finite, smooth outside push is applied.
✓Peak speed becomes unlimited in finite time.
✓Total energy stays bounded.
✓The proof has a public Lean version.
It does not yet establish
×Normal water breaks without a designed push.
×Most fluid flows fail.
×Real liquid literally reaches infinite speed.
×Independent mathematicians broadly accept it.
×Lean alone settles whether the encoded claim is the right one.
08 · Why Mehdi is furious
He is fighting four different battles
Only the first is about the equation itself. The rest are about what counts as understanding, who gets credit, and how science is announced.
10,000agents, about
88hto reach the result
2.7mmessages for this problem
130bnoutput tokens, reported
01
“That is the engineered version.”
He cares about a fluid failing under its own motion, not one steered by a specially chosen push. The distinction is real; his claim that Clay forbids the pushed route is not.
02
“Search is not understanding.”
He sees an industrial search and proof-checking system built on human mathematics—not a machine inventing a fresh way to think about fluids.
03
“The credit story is dirty.”
He alleges OpenAI benefited from private researcher activity and minimized Levent Alpöge. The checked material does not prove this; OpenAI publicly denies seeing the unreleased work.
04
“This is marketing before review.”
He sees the launch as a publicity event for AI capability and valuation, announced before normal independent acceptance. That motive is his interpretation.
09 · The fair verdict
Both sides are talking past each other
The formal question and the physically natural question are related, but they are not identical.
Where OpenAI has a point
The forced route is officially allowed.
Clay’s C and D options explicitly permit a smooth outside force. If the proof is correct and the formal version matches it, this is a serious answer to the written problem.
Where Mehdi has a point
The headline sounds broader than the result.
The theorem does not settle the unforced case people usually picture. Independent review, physical meaning, attribution, and what the AI actually understood remain separate questions.
Bottom line
OpenAI claims one carefully pushed vortex breaks the equations in an officially permitted way. Mehdi objects to treating that narrow, engineered, not-yet-broadly-accepted result as the final conquest of the fluid mystery.
Sources and limits
What was checked
Original criticismhttps://x.com/BetterCallMedhi/status/2097472336257863722
OpenAI announcementopenai.com/index/navier-stokes-solution/
OpenAI proofFinite time blowup for Navier–Stokes (PDF)
Formal versiongithub.com/openai/NavierStokesAndEuler
Official prize wordingClay Mathematics Institute statement (PDF)
Verification boundary: This explainer checks the public claim against the opening theorem, OpenAI’s description, the public repository, and Clay’s wording. It does not independently validate every step of the 165-page proof, rebuild the full Lean project, audit the reported compute, or resolve the competing claims about private data and researcher credit.
AN IDEA TO TAKE WITH YOU
CORE MECHANIC
Neural operators accelerate fluid dynamics predictions by orders of magnitude, but must obey strict energy conservation bounds to handle turbulence.
FIRST-PRINCIPLES LABORATORY · EDITION 001
Explore all studies in the Lab ↗