Nowadays, everybody is talking about LLMs and how they are creating larger and larger language models.
I think this approach is not the right way. From my former experience, I know that sometimes the solution itself already exists, but not in the form we envision. We just need to put the pieces in the right form!
This is the same with Large Language Models. I think from my ethnical and professional background, I try to solve things smart, not hard.
Many people complains they cannot collect sufficient language information for building the large models. Such as my mother language - Hungarian.
These companies solve the problem in the wrong way!
Why the hell do you need Hungarian data to solve a physics, mathematical, or economics problem?
The problem can be solved with an English model, and then it is sufficient to translate the results. If we are honest - we only need Hungarian language model for translation and for historical information about the history of Hungary.