Comment by viccis

Vector spaces and bag of words models are not specifically related to LLMs, so I think that's irrelevant to this topic. It's not about "knowledge", just the ability to represent words in such a way that similarities between them take on useful computational characteristics.