Except these algorithms don’t “know” anything. They convert the data input into a framework to generate (hopefully) sensible text from literal random noise. At no point in that process is knowledge used.