3 ms·
Ask HN: 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.
- PaulHoule 2y agoTabular data is a weak spot for LLMs, probably you are better using algorithms from scikit-learn, probably XGBoost or something similar. There is a big literature in using neural networks together with differential equations and other modeling ideas to do things like predict the weather, understand biological systems, etc. https://link.springer.com/article/10.1007/s10489-023-04824-w https://link.springer.com/article/10.1007/s10489-023-04824-w https://towardsdatascience.com/solving-differential-equations-with-neural-networks-4c6aa7b31c51 https://towardsdatascience.com/solving-differential-equation... are two examples.