Skip to main content
MANIFOLD
[Add answers] Who or what will first solve Navier-Stokes before 2028?
16
Ṁ1.3kṀ5.7k
2027
97%
OpenAI
96%
GPT
3%
an academic mathematician (i.e. currently employed by an academic institution)
2%
Harmonic
2%
GDM
2%
Claude
2%
Axiom Math
1.7%
Anthropic
1.7%
A randomer on Twitter
1%
Grok
1%
xAI
1%
PingYou
1%
Gemini

Resolution criteria

  • What counts as "solved":

    • It doesn't have to be formally accepted by the Clay institute, but it should be generally accepted as clearly correct in essence

    • A lean proof is not required

    • Minor proof errors are accepted, as long as they don't interfere with the consensus belief that the proof is essentially correct

  • Timing

    • If a proof is released before 2028 and it is ambiguous whether it meets this criteria, I may wait arbitrarily until the consensus becomes clear to resolve.

    • A proof that is created before 2028 but not publicly announced until after the deadline does not count.

  • Multiple choice

    • Multiple answers can resolve yes; all qualifying answers count

    • For a company to resolve yes, the solution must be pursued by a company employee as part of company work, or by a model operating broadly under company direction.

      • E.g. a randomer on Twitter who solves it using Claude would resolve Claude yes but would not resolve Anthropic

    • For a model to resolve yes, it must be explicitly clear that the model was used for the solution

      • E.g. GDM announcing the result doesn't automatically resolve Gemini yes, if the result is some specialised internal model.

I won't trade except to set the initial probabilities

Get
Ṁ1,000
to start trading!
Sort by:

We’re sharing a solution to the Navier–Stokes existence and smoothness problem, one of the Millennium Prize Problems. This proof, produced by an internal OpenAI system,

@traders please note the resolution criteria:

For a model to resolve yes, it must be explicitly clear that the model was used for the solution

E.g. GDM announcing the result doesn't automatically resolve Gemini yes, if the result is some specialised internal model.

It's not clear to me at this time that the OpenAI result, assuming it otherwise qualifies, will resolve GPT "yes" also. I'll wait for additional info.

sold Ṁ22 YES

@draaglom 1) All OpenAI models are in the GPT lineage pretty much, and 2) it explicitly says in the news post that GPT Astra was used to create the Lean formalization, which is an important part of the work

@xjp yeah, I just hadn't read the full post yet. Now that I have, I'm inclined to agree that GPT would resolve yes conditional on an overall resolution.