• Home
• Technology

Meta has revealed that its Muse Spark artificial intelligence models have helped mathematicians work on six research problems, including five that had previously remained open without known solutions.
The research covered several areas of mathematics and computational science, including probability, differential equations, group theory, mathematical optimization, computational physics and non-associative algebra.

Muse Spark Takes on Open Mathematical Research

Meta said the experiment differs from Muse Spark’s previous performance in mathematical Olympiad competitions, where the model achieved gold-medal-level results.
Unlike competition problems, open research questions do not have predefined answers or guaranteed methods for reaching a solution.
For the research work, mathematicians used Muse Spark 1.1 and 1.2 in Thinking Mode through the standard Meta AI conversational interface, without relying on a specialized research system or custom modifications to the models.
According to Meta, a team of mathematicians led the research process and developed mathematical arguments in collaboration with the model, while a separate group of researchers reviewed the resulting work.

AI Contributes to Research Across Multiple Mathematical Fields

One of the studies involved probability theory and examined a problem concerning the precise threshold for fitting randomly distributed Gaussian points in high-dimensional spaces inside an ellipsoid.
In differential equations, Muse Spark helped researchers investigate a long-standing question related to wave collapse in a model inspired by laser physics.
The model was also used in group theory, where it generated a research program that identified a counterexample to a mathematical conjecture using a structure of order 384.
Muse Spark additionally contributed to research involving mathematical optimization, computational physics and non-associative algebra, helping researchers develop parts of mathematical arguments and identify counterexamples to certain hypotheses.

Meta Stresses the Role of Human Researchers

Meta emphasized that the results should not be interpreted as evidence that Muse Spark conducted the research independently.
The company said the findings resulted from collaboration between the AI model and human researchers, followed by additional human review.
The research papers distinguish between portions developed by the mathematicians and those generated by the model, while also citing relevant previous research.
Meta also noted that teams outside the company independently announced solutions to some of the same problems using different approaches, with those developments documented in the published papers.

Meta Introduces Open-Source Muse Gadgets Project

Alongside the mathematical research, Meta announced Muse Gadgets, an open-source project designed to allow developers to create physical devices and components that can work with the Muse AI agent.
The project includes ESP32 firmware and a Linux SDK, giving developers tools to build hardware capable of connecting to Muse and carrying out different tasks.
Potential applications include customized E Ink displays, HDMI-based TV connectivity modules and small touchscreen displays similar to the Muse Charm.
Developers can also use commercially available hardware such as ESP32 boards and Raspberry Pi devices to create their own Muse experiences.

Muse Home Link Connects AI With Smart Home Devices

Meta also introduced Muse Home Link, a USB-C device designed to connect Muse with a home network and smart devices.
The system is intended to allow the AI agent to interact with devices such as televisions, speakers and lighting systems.
Meta said it produced an initial batch of 5,000 Muse Home Link devices for distribution to Muse subscribers.
Through the open-source Muse Gadgets project, developers can also build new hardware and experiences around the AI agent.

Meta Expands Muse Beyond Conversational AI

Meta’s latest developments point to two parallel directions for Muse.
The first involves using its AI models for increasingly complex research tasks that go beyond solving problems with known answers.
The second focuses on extending Muse beyond an AI interface by connecting it to physical hardware and enabling the agent to perform tasks in the real world.