On Monday, OpenAI revealed major breakthroughs in mathematics, physics, and computer science, alongside a shifting strategy for sharing its discoveries. However, researchers speaking to The Verge expressed growing exhaustion with the company’s conduct despite its stated course correction.
Recent Breakthroughs and Release Strategies
The three new findings address prominent unsolved problems across different scientific fields, extending past projects and establishing new ground. Following recommendations from appointed “shepherds,” each discovery follows a distinct release path. For the mathematics achievement, OpenAI published the AI-generated output without restriction, allowing researchers to independently organize efforts to explain it. For the computer science finding, the company is forming an external researcher group to provide explanations. Meanwhile, the physics findings remain unreleased, with OpenAI deferring to an external panel of experts regarding how to distribute the materials.
The mathematics achievement claims to resolve the Kakeya conjecture across all dimensions, exploring the minimal space required to rotate a pencil or needle in every direction. Brown University mathematics professor Brendan Hassett told The Verge that the claim appears well supported, though he noted he had not yet verified the Lean formalization used for computational checking. The physics discovery investigates quantum chromodynamics within the Standard Model and its connection to Yang-Mills theory. The final result asserts the disproval of the exponential time hypothesis in theoretical computer science, challenging assumptions regarding how quickly certain problems can be solved.
Criticism Over Handpicked Expert Panels
Experts described the results as massive and significant, but noted that fully evaluating the work takes time due to varying disclosure levels and complexity. Nevertheless, researchers felt frustrated by OpenAI setting up additional barriers through handpicked oversight groups rather than allowing immediate community scrutiny. Constantin Kogler from the Institute for Advanced Study welcomed a more manageable batch of results compared to a previous preprint dump, but questioned why initial access was restricted to a select group of distinguished scientists.
Andreas Karch, a physics professor at the University of Texas at Austin, told The Verge that the handpicked group approach is disastrous, arguing that major AI labs should not dictate winners and losers. Karch suggested that releasing the proof immediately alongside the expert team would democratize access while maintaining meaningful evaluation. This exact form of gatekeeping was cautioned against by the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) just prior to the announcement. The advisory group argued that selecting specific recipients for initial output is unnecessary, risks biased selection and delays, and improperly borrows the reputation of chosen researchers.
Following the new announcements, AGMAI called on OpenAI to immediately publish the natural-language and Lean outputs for the physics findings. The advisory group stated that understanding should progress openly with equal access for all, noting that since the Lean formalization is finished, no scientific reason remains for further delays. Hassett added that open availability fosters better science and accelerates the adoption of new AI tools by global research groups.
Stanford mathematician Jared Duker Lichtman noted that finding a universally satisfying communication method might be impossible, as every strategy involves tradeoffs and potential criticism. Xiaoyu He, a mathematics professor at the Georgia Institute of Technology, suggested the underlying issue goes beyond announcement strategies. He told The Verge that while the results are impressive, the company struggles to address fundamental concerns regarding rapid, non-human solutions and the difficulty of reading and verifying the papers.
