The math community held its breath. It has yet to recover from the catharsis triggered on October 6 by the sudden release of more than 700 mathematical results produced by internal OpenAI frontier models and The next scientific milestone has already arrived. And the artificial intelligence (AI) lab led by Sam Altman claims to have made progress on three very complex scientific problems.
However, on this occasion not all results are strictly linked to research in the field of mathematics. One of them, but the other two address theoretical computer science and physics. In mathematics, the internal AI model used by OpenAI has, in theory, contributed to the clarification of the Kakeya problem, a conjecture that studies the properties of sets that contain a segment of unit length in all directions.
This simple definition is already complicated, it’s true, but what interests us is that Kakaya’s conjecture has engrossed mathematicians for more than three decades. On the other hand, theoretical computing affects field results Estimation of exponential timewhich maintains that, broadly speaking, certain computational problems cannot be solved by exact algorithms in sub-exponential time. OpenAI results in this thesis question.
A third result, from the field of theoretical physics, addresses quantum chromodynamics, the theory that describes the strong interaction between quarks and gluons. Interestingly, in early October when it suddenly released more than 700 results that I mentioned in the first line of this article, OpenAI did not widely publicize these three results. And he didn’t do that because he didn’t want to rekindle the criticism he received from the mathematical community.
OpenAI runs on eggshells
You have no choice. If you don’t want researchers to pounce on you again, you’ll need to find another strategy to publish the results delivered by your sophisticated AI models. Rightly so, OpenAI is looking for the right formula: “Our publication process continues to evolve in response to feedback from the academic community in these disciplines. In our opinion, it is best to treat these manuscripts as raw research data.”
OpenAI confirms that the experts it consulted recommend three different publishing strategies
According to this laboratory, the academic community has the final say on the choice of publication strategy it will use in the future: “To transform this raw research data into scientific knowledge, in this publication We are trying a new approach. “We contacted external experts and asked them to take a coordinating role, organizing teams to understand the new findings.”
Interestingly, OpenAI assures that the experts consulted have recommended three different publication strategies: “For mathematics results, publish the content generated by AI and let researchers organize themselves independently to interpret it (…) For computer science results, form an external group of collaborators to interpret it and, at the same time, anyone can publish it and publish the content by the experimenter. It (…) for physics results, form an external group of collaborators and publish the materials Leave the decision”.
In this situation, it is important that we keep an essential premise in mind: that an AI model provides a formal demonstration does not automatically mean that it has been solved. Scientific problem under discussion. The academic community must verify that the argument is valid, the formalization fits the appropriate statement, and the conclusions withstand expert scrutiny. We don’t yet know if the academy will approve the results that OpenAI’s AI models are delivering, but we probably won’t have to wait long to test it out.
Picture | Jesswin Thomas
More information OpenAI
In Xataka | These two problems have puzzled mathematicians for decades. A genius solved them with the stroke of a pen
