> "matrices"

Saying that AI is "matrices" is like saying human cognition is "neurons." Maybe true at some level, but it's a low-level implementation detail. The important part of a language model is the function that maps tokens to contextual embeddings.