Actually, we should be calling language models libraries
I’ve been meaning to fully write this out for a while, but I can sketch out here:
Language models approximate the usage of language. But a language is only made up of its usage.
A dictionary is a language model, and it implies a library through its references.
A deep learning model approximates the high dimensional possibility state of a multivariable environment.
It searches for meaning in the library of Babel.