GPT-6 Astra helped physicist Nathaniel Craig turn a proposed solution for a long-running particle-physics problem into a detailed mathematical construction. The work addresses how to represent the full Standard Model on a computational lattice, a grid used to simulate fundamental particles and forces.
The difficulty comes from the weak force, which treats left- and right-handed particles differently. Straightforward lattice methods introduce unwanted mirror particles. A technique using overlap fermions can remove them, but researchers had not found a consistent way to set their phase across all required cases for more than 25 years.
Craig, who leads particle theory at UC Santa Barbara and is visiting OpenAI, proposed comparing progressively finer lattices to correct that phase. Astra developed the detailed proof from his strategy. Lattice specialists Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed the result and raised questions that Craig says strengthened confidence in it.
The paper is a preprint, and Craig describes it as a beginning rather than a finished simulation method. The construction still needs broader scrutiny and substantial engineering before it can model realistic physics. The result is therefore evidence that a model can extend a human expert’s technical plan and support a checkable proof, not that AI independently solved the Standard Model or replaced specialist review.