Seems like the right place to ask with ML enthusiasts gathered in one place discussing curves and the things that bend them: what's the thing with potential to

Seems like the right place to ask with ML enthusiasts gathered in one place discussing curves and the things that bend them: what's the thing with potential to obsolete transformers and diffusion models? Is it something old people noticed once LLMs blew up? Something new? Something in-between?