So the humans have the knowledge, and the only way that an LLM can possibly reproduce it is by happenstance after you train it on the tokenized representation of the human knowledge, a process in which it is just as likely to make something up. What is the point of the LLM in this elaborate waste of time? What does it add?

A typewriter prints the letters exactly as you press the corresponding keys. It doesn't invent bullshit on its own as you do. I can't take the comparison to a typwriter as a serious, good faith argument. If you have any worthwhile ideas, please bring those up instead of poorly considered variations on the usual "it's just a tool" platitudes. The comparison is invalid. You are thoughtless for bringing it up.