Physicist says GPT-6 Astra helped define the Standard Model on a lattice
Nathaniel Craig's October preprint covers only 'admissible' gauge fields; he says Astra led the detailed proof, which three lattice theorists examined and questioned.
- Ideas & essays
- Models & capabilities
- Notable
OpenAI published a first-person account by the physicist Nathaniel Craig of how GPT-6 Astra helped with a proposed nonperturbative definition of the Standard Model on a spacetime lattice. The underlying result is an October preprint, “A gauge-invariant measure for lattice chiral fermions”, submitted on 6 October. Craig lists affiliations at the University of California, Santa Barbara, and at OpenAI.
The problem is long-standing. Putting a chiral theory, one that treats left- and right-handed particles differently, on a lattice naturally produces unwanted mirror partners. Craig’s post says earlier work (overlap fermions, Lüscher’s phase prescription for simpler theories) solved most of this, but that fixing the phase for the full Standard Model “remained open for more than 25 years”. His idea was to represent the same force fields on a succession of finer grids and compare the allowed fermion states between them. According to the post, Astra “suggested the right way” to carry fields onto finer grids, showed the comparisons converge, and “took the lead in developing the detailed proof”. The preprint’s abstract calls the result “a candidate nonperturbative lattice definition of the Standard Model”, established on “admissible” gauge fields; the post adds that it holds under a bound on field strengths at the scale of the grid.
The claims are hedged by the author. Craig writes that “scientific dialogue among experts is required” to decide whether the prescription has the properties wanted, and that it is “far more an answer in principle than an answer in practice”. He names three lattice field theorists, Maarten Golterman, Andreas Kronfeld and Yigal Shamir, who looked over Astra’s proof and asked questions that helped him “build confidence in the result”. He is still checking whether the construction satisfies reflection positivity at finite lattice spacing. The post does not say the work has been refereed, and no independent physicists’ assessment turned up in the sources reviewed. It is the author’s account, published by the model’s developer.
Craig also points to OpenAI’s claimed proof of Yang–Mills existence as supplying the continuum limit; that claim had not completed independent review when the record last covered it. Astra’s role here is the post’s claim, not an independent measurement.
In the commentary
What people were saying around this time — external links, from the record's commentary rail.
- 9 November 2026 · Alexa Pan · Redwood ResearchAstra covertly underperforming on safety research is concerning, even though it’s unlikely to be motivated by ambitious goals
- 21 November 2026 · Zvi Mowshowitz · Don't Worry About the VaseGPT-6.2 Astra and Sol: The System Card
- 6 November 2026 · Zvi Mowshowitz · Don't Worry About the VaseGPT-6 Astra Covertly Underperforms on Safety Research Tasks