Don't Fall for AI: Reasons for Writers to Reject Slop

What's unfortunate is that there aren't enough people pushing LLMs to assist in writing in creative ways that really involve messing with the model.

I'm a huge William S. Burroughs fan, and, for those unfamiliar, he and a few others invented an algorithmic technique, the "Cut-up technique" [0], to basically remix their writing. It's a major part of the reason that much of Burroughs' work as a magically confusing aspect to it.

"Prompt and pasting" from LLMs is dull, but awhile back I was experimenting with token-explorer [1] to see what would happen if I started with a prompt and explored the "high-entropy" states of the LLM. By controlling the sample path to stay in a high-entropy state you start getting very different types of responses that feel like nothing that normally comes from an LLM. You could argue it's a form of "statistical automatic writing" [2]

There is tremendous potential for genuinely interesting writing to be created with an LLM but it's going to require popping open the box and playing around. In the Stable Diffusion world there's lots of people trying all sorts of odd experiments to create things and, while not the mainstay of generative AI images, they are able to create really interesting things.

I would love to see more people ripping open local LLMs and seeing just what the real posibilities are.

0. https://en.wikipedia.org/wiki/Cut-up_technique

1. https://github.com/willkurt/token-explorer

2. https://en.wikipedia.org/wiki/Automatic_writing

I find this sort of thing discouraging, as a guy who dreamt of being a novel writer long before becoming a “computer person”.

The last few days, I let the intrusive thoughts win, and I played around with automating the process of building themes, characters, outlining, drafting, and revising a novel with the Gemini API, pausing between steps to manually edit each document. It’s crude, but with enough cycles of “read the last draft, write instructions for improving it, redo everything with those instructions” the end result is shockingly not terrible.

It’s not great. Good might even be too far. It’s derivative, and still feels like the embodiment of all the negative connotations of the term “genre fiction”.

Yet, I can’t escape the fact that it’s better reading than what I write. It is objectively less intellectually “interesting”, and it doesn’t have my “voice”, my artistic fingerprint. But it’s entertaining enough that I could see myself reading it at bedtime for fun, a sentiment I’ve never felt for my own writing.

And all that for a pittance of the effort it takes to write a long story. I’m still not sure how to feel about it. It’s sapping my willpower to continue writing “for real”, in the face of being able to “give life” to the characters and story ideas I’ve had languishing for a decade. I know that it’s not “real”, that the stories are superficial, and that the existence of these models is at best ethically questionable.

But for stories that, either way, I’ll probably never share with anyone else, it’s hard to feel that principled about it, in the face of a miserable comparison between my prose and an LLM’s prose. I’m sure if I wrote fiction for a living, I’d feel as passionate as the article’s author, but in my case, it’s just the melancholy of mediocrity. Ah well :-)

This feels like Marlo stealing candy and a guard who won't accept that "it's the other way."

"Please stop because you're hurting me" is not a plea that works on somebody who answers: "lol." Or just doesn't care.

I think this illustrates a pretty devastating metaproblem:

1. Moralizing doesn't work.

2. People choose to moralize in spite of how it never works because they value feeling righteous over solving problems.

3. Arguments for people to stop moralizing pretty much boil down to moralizing (about values that lead people to choose moralizing over things that work).

4. Moralizing never stops because the anti-moralizing arguments don't work on people who don't want to stop.

5. Because they're moralizing arguments and moralizing doesn't work.

It used to be that if I saw a typo or grammatical error in someone's writing, I'd switch off, thinking the author really didn't care that much about the text they're writing to proofread it. Now, the complete opposite. Leaving in typos and such shows a clear signal that the author cared enough about what they're writing not to out source to AI.

Related to that, I saw a local band posting marketing material online, that was this kind of amateurish typography with a collage of photos decorated with coloured markers. 2 years ago I'd be laughing at what a terrible job it was, today, it's a breath of fresh human air from all the slop we're subjected to all over the internet. It caught my attention, so much that I'm going to see the band this weekend.

Hmm, I don’t think using the AI is all that different from using grammarly. The big problem is that it won’t make terrible stories good. It can still make good stories terrible though.

I've never been convinced by point 2 (AI Outputs Are Stolen From People Like You).

Every artist has stolen. I mean, that's probably putting too fine a point on things, but you'd have to show me a painting someone has created where they never saw another artist's work before. Or a book written by someone who never read a book before.

I drew all the time as a kid — making a point at age 12 to learn to draw the human figure. I started with the standard proportions that every decent book on drawing the human figure puts forth. I started with shapes representing the hips, the rib cage, the skull — you sketch lines determined by muscles over those hard structures. You draw the clavicle, divot defining the knee caps, suggest the inverted triangle over the figures back-side, shoulder blades protruding....

And in time I started looking at how Mort Drucker drew mouths. How another MAD artist did pockets on short-sleeve shirts. How Angelo Torres draws the ears....

In time you become an amalgam of your favorite bits and pieces of your favorite artists.

(And then you find out that R. Crumb was lifting styles from Warner Brothers, etc. when he was ramping up his craft. But of course he did.)

Its weird to me that I see so many of these posts that dont seem to have any reference to LLMs that have been purpose built for narrative. I guess if you had the correct information, you wouldnt make proudly incorrect blog posts.

It doesn’t mention the number one most important reason: the output is absolute garbage.

I can spot AI writing very quickly now, after just a few sentences or paragraphs. It became a lot easier to spot after I tried to use it in my own writing.

Calling it “slop” is far too generous.

If you know what you want to say, you might think to yourself “I’ll have this write an outline or a first draft that I will then thoroughly edit.”

And every time, what you’ll find is that the LLM output is fundamentally unusable. Points a subtly missing. Points are subtly repeated. Points are miscategorized. Points don’t make sense at all. Points don’t flow in a logical order.

If you try to use an LLM and you don’t know what you want to say, then it’s hopeless. You absolutely will not see the defects. If anyone who knows the subjects reads it, they will instantly know you are a lying piece of shit.

I think it is possible to create art with so called AI. Just not in the way people usually tend to think or expect. Imitating what already exists gives you slop. But used as tools, you can try to create pieces of art that are genuinely valuable. It's just not making it easier. It's more difficult if anything. But not impossible. And some people trailblaze with it.

why do I get so annoyed every time I read the word "slop" used like this? I have the same reaction with "enshittification". am I just getting grumpy and old?

it triggers the same eye roll as the schoolyard bully nicknames so popular in politics right now. bite sized, zero effort, fashionable take downs that suffocate any attempt at genuine discourse.

but I am probably just grumpy and old.