OpenAI claims its artificial intelligence systems generated proofs for three major problems in mathematics, theoretical computer science, and physics.
According to reporting from Engadget and The Verge, the artificial intelligence research lab announced a new wave of discoveries across three distinct scientific disciplines. The most prominent finding claims to disprove the Exponential Time Hypothesis, a longstanding assumption in theoretical computer science established in 1999. The hypothesis assumes that certain complex computational problems cannot be solved quickly. If the new subexponential time algorithm generated by OpenAI's model holds up to scrutiny, researchers could theoretically develop far more efficient algorithms for applications spanning databases, geometry, and computational biology.
The computer science findings are being evaluated by an independent working group that includes scientists Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams. Impagliazzo and Paturi originally proposed the hypothesis decades ago. Rather than holding the material in secret, the collaboration group published the artificial intelligence output, Lean-formalized proofs, and supporting code directly to GitHub for public examination.
Academic Partnerships and Multi-Disciplinary Proofs
Beyond computer science, OpenAI released mathematical results establishing Sogge's local smoothing for wave equations across all spatial dimensions, building on earlier research that recently earned NYU professor Hong Wang a Fields Medal for resolving the three-dimensional Kakeya conjecture. For physics, Engadget reported that the company produced an existence proof for interacting quantum field theories containing gauge fields and fermions in four dimensions. A panel of physics experts, led by Nobel laureate David Gross alongside Arthur Jaffe, Michael Douglas, Igor Klebanov, and Nikita Nekrasov, is guiding the verification of those results.
The release strategy represents a shift in how artificial intelligence organizations interact with academia. OpenAI stated that it will not seek formal academic credit or authorship for its models. Instead, external academic collaborators will receive credit for analyzing, contextualizing, and publishing the final manuscripts while ensuring proper citation of existing literature. The Verge noted that an advisory panel of mathematicians recently cautioned AI labs against giving select individuals early exclusive access to outputs, recommending open scholarly repositories instead.
Why It Matters
These announcements highlight a fast-evolving role for artificial intelligence in fundamental scientific research. Rather than acting merely as writing assistants or coding tools, advanced AI models are increasingly capable of proposing novel mathematical steps and candidate proofs for deep theoretical problems. If verified by peer review, disproving foundational concepts like the Exponential Time Hypothesis could unlock breakthroughs in practical software optimization. However, the process also creates new challenges for academia, requiring robust frameworks to verify complex automated proofs, establish fair attribution, and maintain rigorous peer-review standards as machine-generated scientific outputs accelerate.
Illustration: AI-generated image, not a photograph of the event.