Teaching AI to do physics

We build and rigorously test AI systems that carry theoretical physics from a well-posed question to a calculation.

01Benchmarks

NL - Theoretical Physics

02Research log

A lattice calculation, run end to end by an agent

An AI agent ran a first-principles lattice calculation of the SU(3) scalar glueball mass. It wrote its own GPU-accelerated code, blinded itself to the target, and gated its own results, landing on a qualified continuum estimate of M = 2.13(43) GeV.

The log records what the agent did, where the calculation was fragile, and which of its own gates it failed before it reached a number worth quoting.