OpenAI and the Academic Community Explore Hard Problems in Mathematics and Physics

The Academic Community Becomes a Key Link in AI Scientific Verification

OpenAI recently shared its models' results on several fundamental scientific problems, covering the Kakeya conjecture, the Exponential Time Hypothesis (ETH), and Quantum Chromodynamics (QCD). These results are not a simple list but emphasize how the academic community can participate in a verifiable way. According to OpenAI, it collaborated with academic researchers so that model outputs can be independently checked and further built upon.

The core of this approach is open verification: the mathematical derivations or physical inferences produced by the models need to be scrutinized by domain experts. The academic community is no longer a passive recipient of results but an active verifier and extender. This offers a realistic path toward addressing the credibility problem of AI in the hard sciences.

Results and Implications in Three Specific Areas

Kakeya Conjecture: An AI Attempt in Geometric Measure Theory

The Kakeya conjecture is a classic hard problem in geometric measure theory, involving the rotation of line segments within a minimal area. OpenAI shared the model's results on this conjecture but did not claim to directly prove or disprove it. The point is that researchers can verify and carry out follow-up work based on the model outputs. This means the AI's role is closer to proposing testable intermediate steps or constructive ideas rather than replacing mathematicians' rigorous proofs.

The Kakeya conjecture is one of the central problems in geometric measure theory, originating from the "needle problem" proposed by the Japanese mathematician Sōichi Kakeya in 1917: in the plane, what is the smallest area swept out by a unit-length needle as it rotates 360 degrees? Besicovitch later proved that the area can be arbitrarily small, which led to the concept of a Kakeya set—a set containing a unit line segment in every direction. The Kakeya conjecture asserts that in n-dimensional Euclidean space, the Hausdorff dimension of a Kakeya set equals n. The conjecture is deeply connected to restriction problems in harmonic analysis, the Bochner-Riesz conjecture, and other topics, and is a central hard problem at the intersection of geometric measure theory and harmonic analysis. The two-dimensional case was proved by Davies in 1971, but the three-dimensional and higher cases remain unresolved and have long been regarded as a major open problem in the field.

Exponential Time Hypothesis: Cross-Validation in Computational Complexity Theory

The Exponential Time Hypothesis (ETH) is one of the core assumptions in computational complexity theory, concerning algorithmic lower bounds for many problems. OpenAI introduced model results into this area and invited researchers to verify them. Computational complexity theory demands extremely high logical rigor, and any model output must undergo formal verification. Participation by the academic community can help distinguish which outputs have theoretical value and which are merely superficially plausible.

The Exponential Time Hypothesis (ETH) was proposed by Impagliazzo and Paturi in 1999 and is a core assumption in computational complexity theory regarding the complexity of exact algorithms. Its basic form asserts that for the 3-SAT problem, any algorithm has a worst-case time complexity of at least 2^{δn}, where n is the number of variables and δ is some positive constant. The strong form of ETH (SETH) further constrains the complexity of solving the SAT problem. The significance of ETH lies in providing a fine-grained lower-bound framework for a large number of NP-hard problems—if ETH holds, then many problems have no algorithm significantly better than brute-force search. ETH has not yet been proved or disproved, but it has become a foundational assumption in areas such as parameterized complexity and exact algorithm design.

Quantum Chromodynamics: AI Assistance in Physics Computation

Quantum Chromodynamics describes the strong interaction, and its calculations usually involve complex non-perturbative effects. The model results OpenAI shared in this area likewise require verification by physicists. Unlike pure mathematics, physical results must ultimately be checked against experiments or first-principles calculations. Feedback from the academic community can reveal whether model outputs are physically self-consistent and whether they have computable practical value.

Quantum Chromodynamics (QCD) is the fundamental theory describing the strong interaction between quarks and gluons and is a core component of the Standard Model. QCD has two major features: asymptotic freedom and color confinement. Under high-energy conditions the coupling constant becomes small and quarks behave as approximately free particles; under low-energy conditions the coupling constant grows and quarks and gluons are confined within hadrons. Because of color confinement, low-energy QCD calculations cannot use perturbative methods and must rely on non-perturbative techniques such as lattice QCD, solving path integrals numerically on a discrete space-time lattice. Such calculations require large-scale supercomputer resources, and the results must be checked against experimental data (such as the hadron mass spectrum and decay constants), making this one of the most challenging scenarios in AI-assisted physics computation.

Why Verifiability Is the Watershed for AI Scientific Applications

The core signal OpenAI conveyed this time is "verifiability." The value of AI in science lies not only in generating answers but also in generating intermediate products that can be checked. Mathematical proofs can be checked step by step, physical derivations can be compared with known limits, and complexity-theory results can be formally verified.

If model outputs cannot be verified, their significance in the hard sciences is very limited. The involvement of the academic community effectively establishes an external quality-control mechanism for AI results. The speed and rigor of this mechanism will determine the extent to which AI can truly participate in scientific discovery rather than merely producing text that looks professional.

How Researchers Can Participate and Build

OpenAI encourages researchers to verify the model results and continue working on that basis. Specifically, researchers can reproduce the derivation chains given by the models, check for logical gaps in key steps, or use model outputs as heuristic clues and then apply domain methods for rigorous treatment.

This collaborative model is valuable to both sides: AI provides a broad range of candidate ideas, and the academic community provides rigorous screening and deepening. In the long run, this division of labor may be more meaningful than simply pursuing model scores on benchmarks, because it directly targets increments in scientific knowledge.

A Realistic Look at AI Scientific Applications

It is important to remain clear-headed: the model results on the Kakeya conjecture, ETH, and QCD are still at a stage requiring external verification. OpenAI's sharing is more like an invitation than a conclusion. The scientific community's response and follow-up verification work are the key to judging the actual value of these results.

Progress of AI in the hard sciences will not be determined unilaterally by model capabilities alone but depends on the maturity of verification infrastructure and academic collaboration mechanisms. OpenAI's interaction with the academic community this time provides a sample worth observing: when AI begins to venture into the core hard problems of mathematics and physics, open verification may be more important than claims of breakthrough.

One-line Summary

OpenAI shared model results in the Kakeya conjecture, ETH, and QCD, emphasizing that academic-community verification is key to AI scientific applications.

Paragraph Summary

OpenAI recently shared model results in three fundamental scientific fields—the Kakeya conjecture, the Exponential Time Hypothesis (ETH), and Quantum Chromodynamics (QCD)—but did not claim a breakthrough, instead emphasizing independent verification and follow-up construction by the academic community. The article points out that the value of AI in the hard sciences lies not in directly giving answers but in generating checkable intermediate products—mathematical derivations can be checked step by step, physical inferences can be compared with known limits, and complexity-theory results can be formally verified. The academic community thus shifts from passive recipient to active verifier, forming a collaborative model in which AI provides candidate ideas and researchers rigorously screen them. Verifiability is regarded as the watershed for AI scientific applications; the model results are still at a stage awaiting verification, and their actual value depends on the maturity of verification infrastructure and academic collaboration mechanisms.