Executive Summary
OpenAI has announced a significant research breakthrough where its GPT-5.2 model was instrumental in discovering and proving a new formula in theoretical physics. In a preprint co-authored with academic physicists, the research overturns the long-held assumption that "single-minus" gluon tree amplitudes are zero, demonstrating they are non-zero under specific conditions. The model first simplified complex, human-derived equations, then conjectured a general formula from the pattern, which was subsequently proven by an internal version of the model and verified by human experts.
Key Takeaways
* New Physics Result: The research identifies a specific "half-collinear regime" where single-minus gluon tree amplitudes, previously thought to be zero, are in fact non-zero.
* AI-Driven Discovery: The final, simplified formula was first conjectured by GPT-5.2 Pro after it analyzed and greatly simplified complex expressions that human physicists had calculated by hand.
* AI-Assisted Proof: An internal, scaffolded version of GPT-5.2 generated a formal proof for its conjectured formula, which was then analytically verified by the human authors.
* Methodology: The work showcases a new model for AI-assisted science where the AI simplifies complex data, identifies patterns, proposes novel hypotheses (formulas), and aids in the formal proof.
* Generalization: The methodology and insights, with the help of GPT-5.2, have already been extended from gluons to gravitons.
* Availability: The findings are published in a preprint on arXiv titled "Single-minus gluon tree amplitudes are nonzero" and have been submitted for formal publication.
Strategic Importance
This announcement demonstrates the evolution of AI from a tool for analysis to a genuine partner in fundamental scientific discovery, capable of generating novel, verifiable insights that advance the frontiers of complex fields.