AI and machine learning

Physicist credits GPT-6 Astra with leading proof of a lattice Standard Model

Nathaniel Craig of UC Santa Barbara said OpenAI’s GPT-6 Astra took the lead on the detailed proof in his candidate nonperturbative lattice definition of the Standard Model, featured on OpenAI’s new science blog.

Filed under Mathematics, Large language models

Nathaniel Craig, a particle theorist at UC Santa Barbara and a visiting scientist at OpenAI, posted a preprint on 6 October constructing a gauge-invariant measure for overlap fermions on four-dimensional Euclidean lattices, for a class of nonabelian chiral gauge theories that includes the Standard Model. Comparing chiral projectors at successive lattice spacings gives an exponentially local current, which supplies the global properties needed for Martin Lüscher’s reconstruction of a smooth, gauge-invariant fermion measure. The paper comes with a Lean 4 formalisation.

OpenAI presented the result as the first guest post on a science blog it launched on 15 November. Craig wrote that he had a strategy but could not prove it covered every case, and that Astra turned the strategy into a construction and led the detailed proof.

Craig called the result an answer in principle rather than in practice: reflection positivity, a definition in continuous spacetime and realistic simulations remain open. He named three lattice theorists, Maarten Golterman, Andreas Kronfeld and Yigal Shamir, as having reviewed the work; no independent assessment has been published.