Toward Multidisciplinary Scientific Foundation Models

We present the PolymathicAI approach to building foundation models for scientific applications, with a focus on bridging the traditional boundaries between data modalities and scientific domains. See our results on building foundational surrogate models for fluids, trained on a wide diversity of different physical systems, as well as our research on multimodal foundation models to bridge different scientific data types, with a concrete example from the field of astrophysics.