Caltech's Physics AI Ditches Transformers for Neural Operators

Anandkumar and Jenik's new company uses operator learning instead of attention to handle 5 trillion parameters in a single prompt.