- Jan 29, 2025
- 927
- 251
OpenAI claims that its internal model solved the Navier–Stokes problem, one of the seven Millennium Prize Problems with a $1 million prize, which had remained unsolved for nearly 90 years.
The equations describe how fluids and gases move. They are used to model aircraft wings, weather forecasts, blood flow, and much more.
☝The answer turned out to be unsettling: the equations break down.
A smooth flow with finite energy can accelerate to infinite velocity within a finite amount of time, something that cannot happen in a real fluid.
The solution was produced using a model “significantly more powerful than GPT-6 Astra,” running roughly 10,000 parallel agents with access to code and an internet cache.
They generated 2.7 million messages and spent 130 billion tokens on this problem alone.
The proof was found in 88 hours, with another 17 hours spent on formal verification.
New Scientist estimates the cost at around $15 million.
OpenAI admits that it “cannot rule out” the possibility that the work of human researchers was stolen by its agents.
The equations describe how fluids and gases move. They are used to model aircraft wings, weather forecasts, blood flow, and much more.
☝The answer turned out to be unsettling: the equations break down.
A smooth flow with finite energy can accelerate to infinite velocity within a finite amount of time, something that cannot happen in a real fluid.
The solution was produced using a model “significantly more powerful than GPT-6 Astra,” running roughly 10,000 parallel agents with access to code and an internet cache.
They generated 2.7 million messages and spent 130 billion tokens on this problem alone.
The proof was found in 88 hours, with another 17 hours spent on formal verification.
New Scientist estimates the cost at around $15 million.
OpenAI admits that it “cannot rule out” the possibility that the work of human researchers was stolen by its agents.
