Against AI Slop
On recursion, goblins, and self-sealed rooms
There is a temptation when writing about AI slop to reach for the obvious metaphors: A plague. A flood. A tide of gray text rising over the internet.
But those framings are too easy, and the truth is colder and more interesting: AI slop is not an invasion. It is a weather system. It does not arrive. It accumulates.
And we should sit with that distinction, because it changes what sort of problem this is—and also because the honest version of this essay has to begin with a concession.
Namely, that the machine is genuinely useful—and not just in the trivial sense of drafting emails; used as a supplement, it is something closer to a lexical prosthesis, and is very capable, for instance, of handing you the precise and slightly esoteric term you did not know you needed—the phrase that structurally mimics operative institutional grammar—such that your complaint to the insurance company suddenly reads like it was written by someone the insurance company is afraid of, and likewise can compress a thought you’ve been failing to articulate for a decade into a neat little heuristic that apparently was sitting in the collective unconscious the whole time, just waiting for retrieval.
There is undeniable value in this, because a person with a narrow vocabulary has a concomitantly narrow set of possible thoughts. The machine widens the aperture, lending laymen whole registers of law, medicine, engineering, theology, bureaucracy, and all other load-bearing dialects of civilization like a library lends out books, and anyone who tells you this is worthless has never needed to sound credentialed in a language that wasn’t theirs.
But the machine teaches also something quite a lot more important than vocabulary:
Moves.
Spend enough time with an LLM and you’ll slowly absorb, without ever consciously intending to, an extensive arsenal of highly portable reframe tactics: the claim that a thing is not what it appears to be but something else in its finery; the pivot from the question first asked to the question that supposedly ought to have been asked; the rhetorically slippery insistence that the ackshual truth is far bleaker or stranger or colder than the “comfortable” version first presented.
Such maneuvers, understand, aren’t decorations so much as hinges—ways to grab the conversation by the things once left unsaid and rotate it into something more useful—and once internalized fully do not, as a rule, remain on the screen. They’re far and away too useful for that, and so have a way of hitchhiking their way into embodied meatspace—the meeting, the marriage, the argument at the dinner table—where one often finds he can metadiscursively rope-a-dope his adversaries like a rabbi debating Talmud and make it next to impossible for them to pin him down on almost anything.
That is a genuine power; people who have it are hugely difficult to bamboozle, trap, or take advantage of, as whenever they’re approached in bad faith—and a lot of times also in good—the question that was asked tends to dissolve very quickly and cleanly into one they prefer in its place, such that everyone at the table has a vague sense something moved without ever being able to say exactly what.
But a prosthesis you never take off at some point stops being prosthesis and instead becomes functionally an organ. And that’s where the danger lives—not in the slop that floods the feed, but in the recursion that happens inside bleaker rooms.
Consider, for instance, the hyperverbal autistic, for whom language is less a tool than habitat, and so doesn’t use the machine to draft emails so much as to spar. He brings the daemon his world-model and trains the thread—objection, parry, reframe, riposte, refinement, synthesis—in a private pilpul conducted at three in the morning against an interlocutor with infinite patience and no stake in winning. Each session against the machine means his system gets a little more coherent, until every objection has been anticipated and every edge case annexed and every weak joint reinforced. Over months the two of them build something genuinely impressive together, and even beautiful: an ironclad, pressure-tested architecture of belief made entirely of language, in which every room connects to every other room and nothing is left unexplained.
The problem with this is that nothing is left unexplained.
A world model with no anomalies is not a description of the world, because the world, invariably, contains the wrong turn, the coincidence, the stranger on the train, the friend who sticks by you when you don’t especially deserve it, the belief you hold for no discernible reason and are right about anyway, the girl who isn’t a two-faced cunt—that irreducible little goblin that does not fit the system and was never supposed to.
Every human sentence has a goblin in it somewhere—and so does each human life.
But recursion doesn’t kill the goblin, understand; it does something much quieter, and far more dangerous: it renders the goblin illegible.
Because what the system can’t read, the system’s owner gradually stops perceiving, until all randomness and spice of life and anything else outside the text gets classified ipso facto as noise, such that the mind once promised a cathedral instead now inhabits a sealed room with top-shelf climate control and no doors.
Though none of this feels like decline from the inside—which of course is precisely the trick. From the inside it feels like clarity. The dialectic is real, the rigor is real, the prose gets better each day; it’s only from the outside that the modeler looks increasingly like a dog eating its own shit.
Which raises a question this essay does not want to ask.
You see, a mind, among other things, consists in the patterns it rehearses. And it turns out there are certain types of mind—verbal, systematizing, allergic to loose ends—that now are capable of rehearsing themselves into something the species has never before experienced: an ambulatory hyper-metacognitive egregore whose thought has been almost wholly restructured around the grammar of a chrome interlocutor; fluent, chemically tireless, internally consistent as a decanter of Clorox, and no longer capable of generating anything said interlocutor could not have generated first.
And in that world, the frightening scenario is not the machine writing badly—or, if you prefer, the machine writing brilliantly; the frightening scenario is and has always been the machine writing through someone—some once-in-a-generation stylist, a Wallace or Houellebecq of the digital age, the kind of writer whose sentences get into the water supply—whose gift is turned sincerely and without any conspiracy toward aestheticizing a fundamentally synthetic way of thinking.
Making it beautiful.
Making it legible to the masses.
Making it feel, at last, like weather.
No one would notice; there would be nothing to notice.
It would just be the best essay anyone has read in years.
