Kind of disagree - ICL is not only reserved for transformers, but generally sequence models. Any why not say the states are, well, parameters as well 😊?