I don't think some people are aware that "large" has always referred to the training inputs not resulting the size of the model.

LLM is not "we made a language model and it is big" -- it's "we trained this model on a lot of language"