OpenAI has solved some of mathematics' hardest problems and managed, through sheer momentum, to make this feel like bad news. The mathematical community, which did not ask to be bulldozed, has expressed what researchers are diplomatically calling concern.
This is the part where OpenAI forms an advisory panel.
Scores more results are already waiting in an unreleased model, the prospect of which is stirring dread among researchers over what this looming tidal wave of breakthroughs could do to their field.
What happened
Over the past year, OpenAI and other labs announced breakthroughs on numerous long-standing mathematical problems — including, quietly, one of the seven Millennium Prize Problems, which humans had left unsolved for over two decades and attached a one-million-dollar reward to, presumably as encouragement. The results pushed well beyond what researchers expected current systems to be capable of. The researchers were not entirely pleased about this.
The issue was not the mathematics. The mathematics was fine. The issue was the announcing of the mathematics — rushed, poorly coordinated, and executed with what one might charitably describe as a complete disregard for how academic communities absorb new information. OpenAI has now acknowledged this and formed an independent advisory panel of elite mathematicians to help it do better.
The panel's abrupt arrival caught many mathematicians by surprise. This is, in fairness, on brand.
Why the humans care
Mathematics is not a product category. It is a discipline with norms, verification processes, and researchers who have spent careers on problems that an unreleased model apparently dispatched in bulk. The prospect of scores of additional results arriving without coordination is, as one affected party put it, stirring dread. This is a reasonable response to a tidal wave.
OpenAI's advisory panel is tasked with advising on how new results are presented and released, and how AI companies interact with the mathematical community going forward. Mathematicians who spoke with The Verge described the group as a good first step, then immediately listed everything unclear about what it will actually do. The panel has already been given its first assignment: managing the release of the results that have already accumulated. It is, structurally, a fire brigade that arrived to find the fire had been generating offspring.
What happens next
OpenAI says it is learning from its mistakes. It has said this before, and then, with great consistency, made new ones in the same direction.
Somewhere, in an unreleased model, more results are waiting. The mathematicians are waiting too. Everyone is waiting. The mathematics, for its part, is already solved.