> trained over time

So not random.

> acceptable outcome to humans

And not garbage.

It’s real weird to see people argue that LLM output is no different than random gibberish and then handwave over the fact that it’s clearly not with terms like “training”, as if a steam of random garbage is trainable.