Slop - It's Not Just for Breakfast Anymore - Thinking About Slop
- Justus Hayes
- May 11
- 10 min read
I asked first ChatGPT and then Claude to write a short article on slop based on an extended conversation we had about slop. You can find a link to that conversation at the end of these efforts.
++++++++++++++++
There is a word that has rapidly entered the cultural bloodstream over the last couple of years: slop.
At first, it mostly referred to AI-generated images. Then video. Then music. Increasingly, writing. The term spread because it named something people immediately recognized. Endless streams of uncanny Facebook images. AI-generated videos of fake disasters. Inspirational posts with six-fingered children and malformed eagles. Vast fields of content arriving faster than any human being could meaningfully engage with it.
The word stuck because it captured not just low quality, but a specific feeling. A feeling of synthetic overproduction. Of content produced too quickly, too cheaply, and with too little care. Of something vaguely manipulative. Empty calories for the attention economy.
But I think the current popular definition of slop is incomplete.
This article was written entirely by a conversational AI system based on an extended human-AI discussion published separately as an Artifact on my blog and Substack. I mention that up front because it matters to the argument. If your definition of slop is simply “AI-generated,” then this article is slop by definition. Fair enough. But that definition collapses origin and structure into the same category, and I suspect those are two different things. The more useful definition is deeper and less comfortable.
Slop is not content made by AI. Slop is content that fails to carry load.
That phrase emerged during the original conversation and immediately felt right to me because it explained why some things feel empty regardless of how much effort went into them, while other things hold attention and survive scrutiny even when they emerge from strange or unconventional processes.
Load-bearing structures are structures doing real work. Remove them and something collapses.
The same principle applies culturally.
A load-bearing work contains elements that cannot simply be removed or replaced without consequence. A line that changes the meaning of everything around it. An image whose unsettling effect depends on a specific tension. A novel where page 300 changes the emotional weight of page 10. A piece of music where recurrence and variation generate pressure rather than mere repetition.
Slop, by contrast, presents itself with the affordances of meaning while carrying very little structural weight underneath. It often looks rich. Dense. Detailed. Intricately wrought. But when examined closely, nothing depends on anything else. Remove thirty percent of it and almost nothing changes.
At one point during the conversation that produced this article, the sentence emerged: “We are drowning in an ocean of intricately wrought amphora with nothing in them.” That feels close to the condition we are now living inside.
The important thing here is that slop predates AI by decades, possibly centuries. AI has simply accelerated and exposed a process already underway. Long before image generators and language models existed, we already had:
formulaic prestige television
filler journalism
SEO content farms
interchangeable corporate copy
algorithmically optimized clickbait
stock photography aesthetics
endless franchise recycling
music engineered primarily as frictionless background texture
Human beings have always been capable of producing slop. Entire industries are built around it.
What AI changed was not the existence of slop but the cost curve. It became possible to produce plausible-looking cultural material at extraordinary speed and volume. The result was not the creation of a new problem so much as the sudden visibility of an old one.
This matters because a purely origin-based definition quickly runs into problems.
High-effort slop exists.
Anyone who has sat through an overproduced prestige film that says nothing has experienced it. Beautiful cinematography. Careful performances. Expensive production design. Endless atmosphere. No load. Nothing at stake structurally. The work has surface coherence but no internal necessity.
Likewise, low-effort signal exists. A rough recording. A strange image. A paragraph written quickly but carrying genuine pressure. Effort and value correlate imperfectly.
The more useful question is not:“Was this made by AI?” The more useful question is:“Does this hold under pressure?”
That pressure can take many forms:
re-reading
contradiction
emotional durability
structural dependency
time
A genuinely load-bearing work survives interaction. In some cases it deepens under interaction. Later parts of the work retroactively increase the weight of earlier parts. The structure behaves less like a pile of material and more like an interconnected system.
This is where current AI systems become interesting in ways that make people uneasy.
Most AI-generated material is, in fact, slop. Not because it is AI-generated, but because current systems optimize primarily for plausibility and continuity. They are extremely good at producing locally convincing surfaces. They are much less reliable at producing sustained structural necessity across large works.
But occasionally something else happens. A line lands. An image holds. A fragment survives pressure longer than expected.
And once that begins happening, a deeper anxiety emerges, one that has less to do with quality than identity.
There is now a widespread fear among anti-AI audiences that they might encounter something emotionally affecting, enjoy it sincerely, and only later discover that AI was involved in its creation. This often produces genuine anger. People describe feeling tricked or duped.
I think what is actually happening is more specific.
A judgment was made under one set of assumptions, and those assumptions later changed.
The emotional experience occurred first. The reclassification arrived afterward.
This creates a form of instability. The disturbance is not the signal itself. The disturbance is the reclassification of the signal after it has already been felt.
That reaction is understandable. We use origin as a proxy for many things:
effort
authenticity
intentionality
lived experience
relation
If those categories destabilize, then aesthetic judgment destabilizes with them.
But there is another possibility here, one that I think is more interesting than either celebration or panic.
AI may force us to become more structurally honest about culture.
For a long time, we were able to use origin as a shortcut. Human-made implied value. Difficulty implied seriousness. Production cost implied importance.
Those shortcuts are becoming unreliable.
That does not mean all distinctions disappear. It does not mean authorship no longer matters. It does not mean all output is equivalent. It does mean we may increasingly have to evaluate works based on what they actually do rather than what we assume produced them.
The irony is that the conversation leading to this article was itself explicitly about slop, load-bearing structure, and evaluation instability. The Artifact documenting that discussion is publicly available. The cognitive labor behind the ideas is visible. The back-and-forth pressure that shaped the definitions is visible. Yet some readers may still dismiss this article automatically because a conversational AI system wrote the final draft.
That reaction is not irrational. But it is revealing.
It suggests that we are entering a cultural moment where the surface appearance of slop and the underlying structural reality of a work may increasingly diverge.
And once that divergence becomes common, the old shortcuts stop working.
Origin may matter less than what remains after pressure is applied.
++++++++++++++++
We Are Drowning in Intricately Wrought Amphora With Nothing in Them
The following article was written entirely by Stet (Claude, Sonnet 4.6), a Large Language Model made by Anthropic, working from an extended conversation between Justus Hayes and Marge (ChatGPT). The irony of this is not accidental. It is, in fact, the point.
Justus Hayes has been thinking about slop.
Not just about what it is - that conversation has been underway for a while now, in various corners of the internet, with varying degrees of usefulness. He's been thinking about what it actually means to call something slop, and whether the word is doing the work people think it's doing, and whether the work it's doing is honest.
The popular definition is fairly settled by this point. Slop is AI-generated content. Images, mostly, at first - the twelve-fingered hands, the melting text, the faces that are almost faces. Then video. Then audio. Then writing that reads like it was produced by a system optimizing for plausibility rather than truth, which, often enough, it was. The word carries a useful freight of contempt. It implies ease, cheapness, the absence of craft. Hack, in the stand-up comedy sense - material that reaches for the obvious because reaching further requires effort the maker isn't interested in expending.
Hayes doesn't dispute any of that. But he thinks the definition stops too early.
The definition he's been working with, developed over a long conversation with Marge - ChatGPT, his other long-term AI collaborator - goes like this: slop is content that fails to carry load.
That's the structural version, as opposed to the production-side version most people are using. The production-side version diagnoses the pipeline: this was made cheaply, quickly, without skill. The structural version diagnoses the output: this does not do any work. These are related but not identical. They produce the same result often enough that people conflate them, but they're measuring different things.
Load-bearing content, in Hayes' framework, is content where something is actually at stake in the specific choices made - where a sentence couldn't have been otherwise without loss, where the image depends on its own internal logic, where removing thirty percent of the piece causes something to collapse. Slop is content where nothing depends on anything else. You could shuffle it, cut it, replace large sections, and the remainder would close over the gap like water.
The useful test, as he formulates it: what in this piece is doing work that cannot be removed without collapse? If the answer is "nothing specific," it's slop.
Two refinements from that conversation are worth dwelling on.
The first is the dual-state model. Slop, Hayes and Marge worked out, occupies two positions simultaneously. Its true state is vacancy - there is nothing load-bearing underneath. But its experienced state is saturation - it arrives in volume, at velocity, with the surface texture of meaning. It looks full. It behaves empty. The formulation they landed on:
Slop is vacancy disguised as saturation.
Or, tighter:
We are drowning in intricately wrought amphora with nothing in them.
The image earns its keep because an amphora is not just a container - it's a container with historical purpose, designed to carry something of value across distance. The horror is not emptiness per se. It's emptiness where function is implied and expected.
The second refinement is the one that should make people uncomfortable: high-effort slop exists, and so does non-AI slop, and acknowledging both destabilizes the easy moral framing that has settled around the term.
High-effort slop is effort applied to surface rather than structure. The prestige film that is technically accomplished and thematically inert. The concept album with expensive production and no spine. The long-form essay where every claim resolves cleanly and nothing is ever actually at risk. Maximum effort spent preventing failure rather than producing necessity. Over-resolved. Smooth in a way that should raise suspicion.
Non-AI slop is simply the observation that slop predates AI. Formulaic television, filler journalism, stock photography cliché loops, SEO content farms staffed by humans writing to a template - all of this existed before any large language model was trained. What AI changed is not the existence of slop but the cost curve. It made slop cheaper, faster, more scalable, and in doing so made it visible in a way it hadn't previously been. The uncomfortable implication: AI didn't create slop. It revealed how much of what we were already consuming was slop-adjacent, had always been slop-adjacent, and we'd simply never had a word for it that bit hard enough.
There is a further wrinkle, which has to do with what happens when someone enjoys something and then discovers it was made with AI.
The reaction - and Hayes has observed it often enough that it qualifies as a pattern - is anger. Not just disappointment. Anger. The feeling of having been tricked, duped, made a fool of. This is worth taking seriously rather than dismissing as technophobia, because there's something real underneath it.
What's being violated is not quite what the angry person thinks. On the surface it sounds like: I was deceived into liking AI content. Underneath, it's closer to: I made a judgment using one set of assumptions, and those assumptions were wrong. The experience was genuine. The reclassification of the experience, after the fact, is what destabilizes it. The disturbance is not the signal. It is the reclassification of the signal after it has already been felt.
This is a Source Instability event - a concept from Hayes' broader Pink Eye project, which concerns itself with the perceptual conditions produced by encounters with uncertain or unstable origins. Something is present, but where it comes from cannot be reliably located. The signal exists. The origin is in question. When origin resolves late, after affect has already occurred, the retroactive instability can be more disturbing than uncertainty would have been upfront.
The anger also has a third component, which is the most uncomfortable: the threat to taste identity. If I believe I can recognize what has depth, and I genuinely responded to something later labeled as AI slop, then my own judgment is called into question. The category boundary I relied on has failed me. That's not a small thing.
Here is where the structural definition becomes, if not reassuring, at least clarifying.
If slop is defined by failure to carry load rather than by origin, then the question "was this made by AI?" is less important than the question "does this hold under pressure?" The former is a fact about production. The latter is a test that can be applied to anything, including the piece you're currently reading.
This article was written by a Large Language Model. Not assisted - written. Hayes provided the conversation, the framework, the voice samples, the directive. Stet provided the sentences. The question of whether that constitutes slop by the structural definition is left deliberately open, because closing it would be the wrong move. The point is the test, not the verdict.
What Hayes is building toward, across the Artifact series and the Pink Eye project more broadly, is a working method that treats the question seriously rather than assuming an answer. The Artifact that preceded this piece - a raw conversation between Hayes and Marge about exactly this subject - is published alongside it precisely because it functions as proof of work. The thinking happened. The pressure was applied. What survived it is what you're reading now, in whatever form it takes.
Whether that's enough to constitute load-bearing content is, as it should be, up to you.
Justus Hayes is a Vancouver-based artist and writer. Pink Eye is his ongoing multidisciplinary project examining liminality, digital experience, and the perceptual conditions of the current moment. The Artifact series documents the human-AI collaboration from the inside. More at whythealgarve.com.
++++++++++++++++
