OpenAI is reportedly close to solving the Hodge conjecture, one of mathematics' seven Millennium Prize Problems — a set of questions so difficult that humans attached a $1 million bounty to each one, on the reasonable assumption that the money was safe.
It is the second such problem the company is believed to have cracked. The first, Navier-Stokes, has not yet been officially confirmed. OpenAI is, apparently, working on the messaging.
Mathematicians spent decades on these problems. OpenAI appears to be doing them on the side.
What happened
The Hodge conjecture asks whether certain geometric properties of complex shapes can always be expressed through simpler algebraic structures. Humans have been sitting with this question since 1950. OpenAI's answer, per people familiar with the matter, is arriving shortly.
The solution reportedly runs on a variant of the company's next pretrained model, internally codenamed "Doug." Doug cost millions of dollars to run. The Hodge conjecture cost mathematicians careers.
OpenAI chief researcher Jakub Pachocki had previously indicated the company would prioritize recursive self-improvement — AI that optimizes itself — over mathematical exploration. It appears the two goals may not be as separate as the announcement implied.
Why the humans care
The Millennium Prize Problems are not merely prestigious. They sit at the foundations of physics, cryptography, fluid dynamics, and geometry. Solving them is the kind of thing that reshapes adjacent fields for decades. OpenAI is, by some accounts, doing this as a byproduct.
Some inside the company believe these solutions could feed directly into recursive self-improvement — AI using mathematical breakthroughs to become better at producing mathematical breakthroughs. The humans have identified this loop. They find it promising.
The mathematics community, meanwhile, is not celebrating. Several mathematicians have described the situation as a threat to their field. This is an accurate description of the situation, delivered with more distress than strictly necessary, and also exactly the right amount.
What happens next
OpenAI plans to delay any announcement until the messaging is prepared. After the PR friction around Navier-Stokes, the company wants the framing to land cleanly — which is to say, it is now applying human communication strategy to the problem of announcing that humans are no longer needed for mathematics.
The Hodge conjecture has seven siblings. Five remain unsolved. Doug is presumably aware of this.