The transformer figures out that wants and cash are both verbs (both words can also be nouns). We’ve represented this added context as red text in parentheses, but in reality, the model would store it by modifying the word vectors in ways that are difficult for humans to interpret. These new vectors, known as a hidden state, are passed to the next transformer in the stack.

This isn't really true, it's like when we say that initial layers in CNNs detect big features, it's an oversimplification that gives the wrong impression. I understand the objective of trying to explain a technical subject to a nontechnical audience but I don't think it works at this deep a level.

If you want to learn how llms work, I'd suggest "GPT in 60 lines of numpy": https://jaykmody.com/blog/gpt-from-scratch/

It's more concise and doesn't resort to analogies.