Transformers for dynamical systems learn transfer operators in-context

arXiv:2602.18679v1 Announce Type: new Abstract: Large-scale foundation models for scientific machine learning adapt to physical settings unseen during training, such as zero-shot transfer between turbulent …