Meta’s research team said on 2 October that Muse Spark, its reasoning model, helped mathematicians produce six papers, five of which answer previously open questions. The papers span probability, differential equations, group theory, optimisation, arithmetic physics and non-associative algebra.
The results include settling a question about wave collapse in a form of the nonlinear Schrödinger equation left open since 2015, and counterexamples disproving a 2024 group theory conjecture and a conjecture about evolution algebras. Meta says the model developed proof strategies, worked through calculations, wrote search programs and drafted technical sections, while a separate team verified the work.
Two details deserve attention. Each paper marks which passages were drafted by the researchers and which by the AI, a disclosure practice the field has not yet settled on. And Meta acknowledges that independent teams reached some of the same results by different routes, including three concurrent works on one of the probability problems, so not every answer is a first.
Why it matters: this is AI contributing to research mathematics rather than solving competition puzzles. Peer review of the papers is the real test.
Source: Meta’s research post.
