Then where does the knowledge come from

How did all those parameters arrive at:

Paris

Apparently, the billions of parameters had been adjusted countless times during training, using vast amounts of text. Books, articles, reference works.

When asked:

What is the capital of France?

the question was converted into the starting numbers. Those numbers then passed through billions of adjusted parameters, determining what was likely to come next.

In this example, the response began with:

With each new word added, the expanded context passed through the billions of adjusted parameters again, determining the next likely word.

The capital of France is Paris.

Somehow, the answer came from completing the sentence. In the language of machine learning, this is called prediction.

Resolved It was really prediction all along.