Hi, Max! Do you have an “easy to explain at a dinner party” definition of AI slop?

I just call it weird fake AI crap. But the phrase that the AI world likes is “unwanted AI-generated content.” The idea being that with slop, it’s like AI shit sort of inserted into places that you didn’t actually want to have it. When you’re talking to a regular person and they’re just thinking about their encounters with it, it’s the weirdness and the fakeness and the incorrectness of it. Most AI stuff is kind of unwanted right now. The Google search answers to me are very unwanted, and some of them are pure slop in the sense that they’re just wrong and weird.

Unwanted is a great word for this. And we usually use that word to refer to pregnancies, so that’s fine. Do you collect slop? Have you been screenshotting slop for years? 

I like weird internet things. It’s one of the beats that I’ve always written about. I have a Google sheet of weird Facebook pages I’ve found. I also asked my newsletter readers to submit examples that they found.

I feel like you have some lurking thoughts about models of payment for slop. Facebook, TikTok, and X are all offering payment for “engaging content.” But for instance, with Spotify, you note, we call this “royalties.”

All the slop is kind of the same, but it’s incentivized in slightly different ways depending on the platform that it comes from. Anybody who uses Twitter has a direct experience of what happens when you start to pay people for engagement  — you get crap, and it’s weird crap because that’s the stuff that gets engagement. It’s offensive, it’s disgusting, it’s emotionally charged. Whereas Amazon, you’re kind of doing a more traditional SEO scam. I talked to a librarian who was telling me about the AI books that are now relatively consistently crossing his desk, requested by patrons and by member libraries. They’re about any kind of diet, any major sort of illness. If you get a diagnosis of fibromyalgia, you’re going to go online, and you’re going to search books about fibromyalgia, and you might ask your library to order a book about fibromyalgia, or you might just order it yourself from Amazon. And there are hundreds of sloppers or scammers, hustlers, guys who are getting ChatGPT to write a book called How to Live With Fibromyalgia or whatever. It doesn’t take them any work to get the book made. The work they do is the SEO work of getting that to the top of the Amazon results. The Spotify thing is funny because royalties are the exact same model as Facebook content payment. It’s just that nobody’s ever had to think about algorithmic juking of royalties before. Actually at Spotify, what you’re doing is not really making engaging music. You’re making music that might get inserted onto a playlist that billions of people are listening to.

Another exciting element you’ve seen is the intrusion, I guess, of developing-world players directly into our ecosystem. We’ve sort of eliminated the native-English-speaking overlords who were using inexpensive labor in other parts of the world. Part of the uncanny of AI here feels like it’s people using English as a second or 10th language and that going through a machine. How much of AI weirdness is internationalism and also people trying to interpret America back to us?

A few months ago, there was a set of recipe videos going viral because somebody had suggested they were all created by AI — throwing a bunch of whole limes and a ton of condensed milk and ice into a blender and calling it lemonade or whatever. But it’s not AI. What it is is …  just Brazilian. Brazilians use a ton of condensed milk in their food. And Brazilians love Brazilian lemonade, which is just limes and milk and condensed milk and ice. And my joke is: Anytime you see something weird online, you have to ask yourself, Is this AI or is it just Brazilian? Am I encountering something like the bizarre product of a large language model, or is it just a cultural practice that I had no idea about?

This has a converse. There’s a host of YouTube videos online, mostly made for Indian people, to take advantage of the Facebook bonus creator program. Jason Koebler at 404 media did a big survey of these videos; he has all these instances of these sort of Indian hustle-culture guys speaking in front of a whiteboard or whatever, being like “Americans love pets.” One of the Facebook guys I talked to said that the big things that he likes to do are pictures about things Americans love — the U.S. military, pets, Jesus.

But it’s much less common than you might expect that it’s the LLM or ChatGPT that is inserting the bizarreness. A lot of the guys who are making this stuff, they’re intentionally making it as weird as possible. That’s what gets you engagement. So you might imagine that the LLM is inserting the visual psychosis of a given slop image. But in fact it’s that the guys on the end are like, No, we got to bump up the weirdness of this one.

As far as the hustlers online, they’re the main characters just trying to get money off of the Americans who are maybe too stupid or too crazy to realize what they’re seeing or who are just willing to share stuff. Perhaps it’s quite American of me to even admire these hustlers for seeing a way to make a quick buck pretty easily and taking advantage of people, even if it has the long-term broad effect of ruining the entire information economy.

I think I agree with you that in most minor ways this stuff is harmless. Facebook bait doesn’t really have a victim. We don’t talk about politics or the election here at all, which seems like that is a slop that’s made by people who have an agenda. They’re not doing that for the money — well, maybe they are — but they’re also doing that to undermine an election or to engineer an election result.

To me, this is the whole Republican Party right now. Sure, there is an ideological basis, but the ideological basis is always subject to the moneymaking imperative. So you have guys who are Trump voters who are hiring Filipinos to write pro-Trump blogs for them, and then also using those as content farms on Facebook. But there’s not a lot of slop being created with the intent to mislead people. There’s a lot more kind of propagandistic slop. The week that everybody was talking about Haitian immigrants in Springfield, a bunch of right-wingers, including Donald Trump Jr., were just creating all these images of Donald Trump protecting cats with guns and stuff.

They were so good!

Yeah, it’s this bizarre sort of kitsch that seems particularly — I don’t have the pretension or the intellectual capability to theorize this — but it seems particularly beloved by fascists. The far-right German party has all these ads that are made using AI. There’s something about the way that the AI’s default to cliché seems particularly appealing to the kind of proto-fascism or, whatever you want to call it, fascist-adjacent politics that we have today. The sort of fake “Make America Great Again.”

The platform you don’t discuss here is Reddit, which is currently overrun with a couple things. One is people manufacturing or using tools to form questions and answers, and then also just tons of insane machine-made storytelling. The vast majority of shit you see on Reddit now is absolutely invented machine stuff, and it seems really disastrous to the platform in some way that I don’t understand yet.

Everybody adds Reddit to the end of their Google queries because that’s where you find actual humans. So obviously you’re engaged in an arms race against SEO people who want to push their links and their whatever else. They all realized that if they can start a Reddit account and stick their Reddit stuff in the top Google rank search for the dashboard lights that are lighting up when you start up your car or whatever, and they can get their auto-mechanic site or whatever into that Reddit thread, that that’s worthwhile to them. It’s quite bad for Reddit because what Reddit has going for it is that it’s one of the last places online where you can consistently encounter humans and human opinions and human stories. It’s quite bad for it to get infected in the same way.

My takeaway is that, in the end, you actually don’t seem too worried. We know that Google is ruined, and the internet is a desolate landscape of falsehoods, which is terrible in an age when people won’t vaccinate their children and then the internet finds a way to reassure them that they shouldn’t vaccinate their children. But in the end, you seem not too worried about the rise of slop.

I don’t think that AI is necessarily making the world worse to a degree that it requires a different kind of attention. Slop is less like a new challenge that we’re confronting and more the pinnacle of a challenge we’ve already been confronting. That’s since platforms first took over the entire media and politics and entertainment and sports industries. I’m really wary of framings of this problem that buy into the self-conception of AI geniuses as creating a brand-new, totally unprecedented, insane world-changing software. I say this even as somebody who is extraordinarily impressed by the capabilities of generative AI. But the current situation with the way platforms process and mediate basically all information passing between people? That just seems unconscionably bad to me. It has been bad. It continues to be bad. One thing I think the thing that really needs to happen with respect to LLMs is that institutions like libraries — let’s call them gatekeeping institutions, even though we hate gatekeepers — institutions like libraries, they need to be funded and given the ability to process and deal with what’s happening here

Neil Clark of Clarkesworld, the small SF magazine, was getting so many AI submissions driven to them by all these YouTubers selling get rich quick schemes. But Neil said the real threat to his business last year wasn’t the slop. The real threat was that Amazon stopped offering subscriptions to Clarkesworld through its Kindle subscription service. So the actual existential threat to him is not AI and slop; it’s that a company a thousand times the size of his is making decisions on a whim that will completely affect your ability to continue your business.

I think teachers are besieged by slop and they have some tools but not totally effective tools. They aren’t keeping pace. I think even they would agree with you. But I feel for people who are on the front lines facing down the slop. 

Totally. Teachers in particular are going to be affected so much more than the average person. And I don’t envy being a first-year writing professor at a college right now. I heard of one teacher who switched everything to in-class and then the students only use ChatGPT to study, but still everything they wrote still sounds like ChatGPT. That stuff does feel quite bad, and I really hope that it’s something that we can metabolize as a society and change it, rather than just something we accept as the new norm.

I actually asked ChatGPT for questions for you. They weren’t awful, but they were just sad. Sort of pitiful. 

I don’t know that this is a solution, but I’m in favor of no guardrails on LLMs. I think that they should be letting this stuff be available to us in the fullest weirdness possible. Then at least if everybody’s going to be using it, it’s weirder and stranger and wrong more often than it is. That’s instead of working toward the most bland and clichéd version of language and imagery possible. If we’re all going to hell in a handbasket, let’s make it beautiful and dreamlike and amazing.

(Is this AI slop? No! These are made by hand, by me!)

You saw the whole scammy Bridgerton party thing with the pole dancer, right? It’s time to retire the phrase “lots going on here” but … in this case?

Thanks for the help, evil robots!

Sign up for The City Desk, a weekly newsletter about New York.