Following the release of hundreds of manuscripts addressing long-standing mathematical questions, OpenAI has now unveiled three additional major scientific findings. This time, the company is collaborating with external experts, referred to as 'shepherds,' to guide the release of results in computer science and physics. For the mathematical breakthrough, OpenAI has opted to share the AI-generated output directly, allowing researchers to explore their own explanations.
One of the discoveries challenges the Exponential Time Hypothesis (ETH), which has long been considered a foundational assumption in computational complexity. The paper presents a deterministic algorithm that refutes ETH by solving a specific case of 3-SAT in subexponential time. An independent group of computer scientists, including Russell Impagliazzo and Ryan Williams, will oversee the verification of this result. OpenAI has made the materials publicly available to ensure transparency and peer review.
The company acknowledges potential issues with its current approach, citing concerns from its advisory panel about unequal access and biased selection. The panel recommended that AI labs release results on scholarly repositories with detailed documentation, funding for researchers, and other support mechanisms. The physics findings, while not fully released, are described as a rigorous existence proof of certain quantum field theories, including Quantum Chromodynamics (QCD) and a unified completion of the Standard Model.
OpenAI has also released an open letter seeking community input on how to handle AI-generated scientific breakthroughs. The letter raises key questions about early access to results, credit recognition, and support for researchers as methods evolve. These developments mark a significant shift in how AI is being integrated into scientific research, with a focus on transparency, collaboration, and long-term academic engagement.