And this training data somehow includes an explanation of how these individual lines (with variable names unique to my application) work together in my unique combination to produce a very specific result? I don't buy it.

And...

> pretty much

Is it "pretty much" or "all"? The claim that the LLM simply has simply memorized all of its responses seems to require "all."