Meta Says Muse Spark Helped Answer Five Open Math Questions Across Six Papers

Meta said on October 2 that mathematicians working with Muse Spark produced six research papers, five of which address previously open questions. The release describes human-guided research with AI assistance: mathematicians chose and developed the work, and another group reviewed it. It is not a claim that a chatbot independently solved six problems. 17

What the six papers cover

The papers range across several areas of mathematics. Examples of the reported work include:

• Probability and geometry: The Strict Threshold for Gaussian Ellipsoid Fitting establishes a threshold around (n \sim d^2/4) for whether a positive-semidefinite matrix can satisfy constraints defined by random Gaussian vectors. 1
• Group theory: one result gives a 384-element group that serves as a counterexample to a conjecture proposed in 2024. 20
• Algebra: On Solvable Evolution Algebras and a Conjecture by García-Martínez and Pérez-Rodríguez addresses the named conjecture. 4
• Optimization: Tightness of the Cycle-Based Relaxation for Completed Length-Three Alpha-Cycles examines whether a particular cycle-based relaxation is tight. 4
• Mathematical physics: String Two-Point Function = Height Function on a Curve is among the listed publications. 4
• Differential equations: the publication list also includes work on finite-time blow-up for radial negative-energy solutions of a mass-critical biharmonic nonlinear Schrödinger equation.

Meta’s headline count should be read carefully: it says five of the six papers answer previously open questions. The available paper listings do not provide a clear, independently verified one-to-one mapping between all six papers and those five questions. 174

How mathematicians worked with Muse Spark

Researchers used Muse Spark 1.1 and 1.2 in Thinking Mode through the regular meta.ai chat interface, without a custom research scaffold. Meta says mathematicians guided the research and developed arguments with the model; a second group of mathematicians reviewed the work. The papers also identify passages primarily drafted by researchers versus passages drafted with AI. 17

That process matters when interpreting the results. A model can contribute ideas, arguments or code, but the research account assigns problem selection and review to people. In the group-theory example, outside reporting says Muse Spark wrote a search program for GAP; mathematicians checked the counterexample and completed the argument. 20

What remains uncertain

The papers are evidence that an AI model can contribute to supervised mathematical research, but they do not by themselves establish that the results are exclusive or that the model’s mathematical output is generally reliable. Other reporting describes several results as having been reached independently by other teams, so the degree of overlap matters when discussing priority.

Mathematician review is also distinct from formal verification, in which a proof is encoded in a strict language so a computer can check it step by step. Meta’s description discusses human review, but the available material does not establish whether each paper’s proof was machine-checked in a proof assistant. 17

Muse Spark was introduced on April 8, 2026, as the first model from Meta Superintelligence Labs. The October papers show a later use of the model in a research collaboration, but they should not be taken as proof that AI-generated mathematical claims are error-free. Assessing that requires examining the papers’ arguments and their subsequent independent verification.