đŹâWe have foundation models for language, not for physicsâ â Anima Anandkumar, Bren Professor of Computing
Article image or reusable cover for Latent Space
Anima Anandkumar, a professor at Caltech, has developed AI models to simulate physical systems such as weather, fusion, and fluid flowâareas that traditionally required massive supercomputers and decades of development.
Her technique, Neural Operators, allows AI to learn from both data and physical laws simultaneously, enabling models to work with much smaller datasets than conventional language models require. She successfully created FourCastNet, which can predict weather as accurately as classical physics-based simulations but runs on standard graphics cards. The key is to build in the structure that the physical world already hasâfor example, by using the correct mathematical foundation for a problemâinstead of trying to train on impossibly large datasets.
Vibekollen prepared this summary with AI from the original publication. The content belongs to Latent Space.
