Are NLP and LLM's suited for non-text based input?

I am reading a book on NLP and LLM's and I am in doubt.

To me it seems that the prediction of anything in NLP and LLMs is built around text.

Text as in not numbers. I am well aware that you treat numbers as text but that's not really what I believe is how you should look at it ?

Let's say you want to predict traffic routing. That's not text so is it merely a different architecture you use for such tasks ?

Let's say you want to predict the weather. Not text. Then what ?

I am generally curious about where the strong sides lie besides the use of LMM's and the underlying architecture for text.

Not everything is text based. Lots of things is a numbers game.