It's quite possible a human forms a full representation of a coherent thought before translating that thought into words, meaning no token by token prediction.
Right, but that would still be functionally quite different from what an LLM does, given that it has no full thought and is entirely predictive by token.
And digital audio will never match analog. Can anybody tell the difference anymore?
If a machine produces highly similar output does it matter? LLM's clearly exhibit behavior of having implicitly learned systems, which is the valuable part of intelligence. Humans infer systems through text too, by the way.
I’m surely not the only one who sometimes can’t “find the right word” in the middle of a sentence when trying to describe a thought or idea.