Post

The AI booster argument that "oh well, human writers are also influenced by other writers" is both insanely dumb and insanely insulting; kind of thing that should get you branded for life as a dolt.

LLMs are token-processing engines. *all they do* is take in tokenized input and spit out output. I say "input and output" because it's not even *language,* the *language* can only happen through human interpretation of those inputs and outputs, externally to the model.

Even if you're a delusional LLM-consciousness maximalist you have to concede that an LLM exists purely in a world of language. A human writer has a huge continuum of experiences that are *inaccessible through the pure ingestion of language-like symbols.*

That should be a very obvious and evident thing to observe! A human writer is not sourcing the entirety of what goes into their writing from other writing that they've read.

And furthermore, when we talk about writers being "influenced" by the work of other writers, we are describing a process that is in no way analogous to generative model training, and to suggest that it is is just shows that you don't know piss about the subject matter here.

It's like thinking that reading is equivalent to eating the glue off the spines of books. It's not! It's just not.

An LLM cannot be said to be "influenced" by a text any more that the blueberries "influenced" my morning smoothie. It just has the text in its dataset.

You have to be just a complete dumbass who engages in the most rudimentary category errors to say shit like this. You have to have a model of how people think that is so grossly simplistic that you think a person reading a book and my computer gzipping a text file is the same thing.

If you have these idiot machines that you are thoroughly convinced are people or, more likely, gods—what are we to do with you losers? All your arguments are attempts at dragging us down to their, and frankly your, level. Of course we think you're hogs and we want to kill you with hammers.