The language model can write like a human, but it doesn’t have a clue what it’s saying.