Cy on Nostr: They pick the next word according to what the most likely next word is. With some ...

They pick the next word according to what the most likely next word is. With some randomness thrown in to keep them from going into infinite loops. That's it. The only distinction is they can guess relatively well according to what the past several dozen (hundred?) words are. They're still guessing based on nothing but randomness, and their training data, which are people's words.

It really becomes apparent when you ask them to write something with a lot of words, like a book. But even "press releases" are long enough it's pretty obvious. And they're only repeating what people have already said, just randomly mixed up to give you the illusion that it's something new, by hiding their sources.