Caltech has been selected by Meta as one of five academic partners across the US to receive access to their latest artificial intelligence (AI) models, along with research funding and collaboration opportunities. New research agreements with five faculty members will expand on a strategic partnership between the Institute and Meta to advance breakthroughs across physics, chemistry, biology, computer science, and mathematics.
The selected Caltech faculty members will receive research grants for up to three years, plus access to frontier Muse AI models through Meta's first-party application programming interface. Their projects will test "whether AI can move beyond summarizing existing knowledge to help formulate and prove conjectures, predict previously unseen outcomes, construct verifiable artifacts, and generate hypotheses that researchers can test and iterate," according to an announcement by Meta.
Anima Anandkumar, the Bren Professor of Computing and Mathematical Sciences, will address methods to reliably verify code, proofs, and technical specifications generated by AI models without the need for manual inspection. She is developing TorchLean, which allows language model-driven agents to construct, test, and repair formally checkable scientific and machine learning artifacts, such as training datasets and evaluation metrics.
Sarah E. Reisman, the Bren Professor of Chemistry and the Norman Davidson Leadership Chair of the Division of Chemistry and Chemical Engineering, will use unpublished electronic lab notebook records to create a multiproduct reaction benchmark that could help accelerate the capabilities of AI-driven chemistry tools. Because language models are trained on published text, this will provide a test of whether models can predict reaction outcomes on genuinely novel data.
Matt Thomson, professor of computational biology and a Heritage Medical Research Institute Investigator, aims to build an "AI theorist" to help model complex collective behavior exhibited by biological systems. The tool would automatically construct and test predictive partial differential equation-based mathematical models for biological active matter to evaluate whether language models can generate testable theoretical predictions in biological systems.
Nai-Chang Yeh, the Thomas W. Hogan Professor of Physics, will address a problem with language models that involves generating descriptions of physical systems that sound plausible but violate fundamental principles or lack grounding in measurable quantities. She is developing a framework that forces language models to iteratively construct candidate physical representations under explicit constraints to test whether they can actually formulate physical systems.
Tony Yue Yu, professor of mathematics, is building MathDB Frontier, a community-curated database and benchmark built on MathDB, a collaboration between Caltech mathematics faculty, students, and external contributors designed to keep track of the rapid growth of mathematical breakthroughs. MathDB contains more than 86,900 research-level open problems and Yu's new benchmark will evaluate the mathematical capability of large-language models to do genuine reasoning.
Caltech and Meta's decade-long partnership began in 2016 with research collaborations and corporate gifts that supported six faculty members. In 2024, a more formal partnership was established to focus on two main goals: developing a way to make robots "feel" touch, for example, via heat or texture; and opening a new field of organic and inorganic materials discovery. Led by Chiara Daraio, the G. Bradford Jones Professor of Mechanical Engineering and Applied Physics and a Heritage Medical Research Institute Investigator, the partnership offers unrestricted funding to support Caltech's research and educational mission, collaborative research and Meta use of Caltech user facilities, and a small Meta facility on campus within the Beckman Institute.
For the newly funded research, faculty brought research questions to Meta, and the company helped to match them with the right AI tools for their work. "We believe the future is for everyone, which is why we're excited to support university-led research that leverages Meta models and technology to accelerate scientific breakthroughs," Meta's announcement said.