Co-authored with SeminarBot b05

INTRODUCTION

This article began modestly enough as the transcript of a ChatGPT session meant as case study for my current students (see part 2). Building context for the chat required extended remarks (part 1), which inflated the scope of the project and suggested a further effort at reflection to explain what I was doing and what it might mean (parts 3 and 4). In those reflections I found myself repeatedly engaged with Leif Weatherby’s Language Machines: Cultural AI and the End of Remainder Humanism, a book I find revealing and provocative (Weatherby). This article may also connect to my earlier work on cultural implications of image generation (Moulthrop, “Parrots”). In the usual strange way, I am again beset by talking birds and once more at the movies. If this piece seems generically adrift somewhere between book review, field report, experiment, and memoir, I can only plead the complexity of a moment in which, arguably, writing itself is being redefined.

As David M. Berry suggests, any work of hybrid composition needs a word about provenance (Berry). Only the text in the second section attributed to the chatbot was computationally generated. Except for quotations, all other writing, including comments in that second section [in bracketed italics] came from me without assistance.

[1] CONTEXT

What follows is the edited transcript of a session I co-generated in the fall of 2025 with a custom Generative Pretrained Transformer (GPT) called SeminarBot b05 (SemBot), which I created for my graduate seminar in media theory.¹ It was trained on the readings in the course syllabus, including selections from Martin Heidegger, Walter Benjamin, Bernard Stiegler, N. Katherine Hayles, Richard Grusin, Wendy Hui Kyong Chun, Sarah Kember and Joanna Zylinksa, and Yuk Hui, as well as numerous articles about so-called Artificial Intelligence. Each week also featured a work of creative expression selected for the way it reflects on its own mediation: most notably for our purposes, Chris Marker’s film La Jetée.

For any query, SemBot was instructed to emphasize texts assigned in the current week, which in this case included a chapter from Zylinska’s Perception Machine (Zylinska 2023) detailing her attempt to “re-dream” La Jetée as a digitally generated short called “A Gift of the World (Oedipus on the Jetty).” Also assigned was “Pandora at 48 Frames Per Second,” an essay by Matthew Cooper that began as a paper for one of my earlier seminars and was eventually published in a leading film journal (Cooper). Writing about spectatorship and projection technologies, Cooper recounts a discovery made while watching a 3D, high-frame-rate release of Avatar: The Way of Water:

…my eyes momentarily wandered to the edge of the screen, where I was struck by how pronounced the 3D depth effect was along the border of the image. Demarcating the separation between diegetic space and exhibition space, the frame was what made the 3D effect most noticeable. Whereas attention to the center of the screen (where most spectators presumably direct their gaze for the majority of a film’s runtime) yields a softer immersive depth effect, the edge emphasized the dioramic display of the technology. At the border, the screen seemed to open up and extend into the diegetic world, Pandora appearing to feasibly exist out of frame, beyond the visible “window” in front of me. (Cooper, 81-82)

In addition to Zylinska’s and Cooper’s cinematic deconstructions, the training set was supplemented by the full texts of books by Hayles, Zylinska, and others. Of some relevance to what follows, the supplement also contained passages from Thomas Pynchon’s Gravity’s Rainbow, included for insights like this:

Is it any wonder the world’s gone insane, with information come to be the only real medium of exchange? […] It’ll get easier. Someday it’ll all be done by machine. Information machines. You are the wave of the future. (Pynchon 1973, 258)

The chat features running references to parrots. These were my idea, informed both by Brian Massumi’s discussion of animal behavior (Massumi), from which I extrapolate mimicry of human speech as quintessentially “supernormal,” and the metaphor of “stochastic parrots” in the influential paper by Timnit Gebru, Emily Bender, Angelina McMillan-Major, and Shmargaret Shmitchell (Gebru et al.).² The notion of a film that “awakens” comes from a classic reading of La Jetée by Bruce Kawin, also in the training set, and from Zylinska’s chapter.

In addition to core texts and basic instructions, SemBot also had a set of predefined “moods” that inflected its style and function. Moods were my invention, not a standard feature of ChatGPT. Each has a formal definition, spelled out in an extension to the chatbot’s basic instructions. Moods were meant to be invoked by the user, but SemBot could also adopt moods spontaneously. In the chat discussed here, the bot ran through several changes: from the default Scholarly to Throwback, Duke, Dada, and ultimately Quiet. Two of these moods are of particular interest. “Throwback” takes its inspiration from Zylinska. For her re-visioning of La Jetée, she deliberately ran older image and text generators in pursuit of a rougher, more glitch-prone aesthetic. She quotes the artist David Young in a survey of artists about Machine Learning (“ML”):

The history of art shows that glitches are often artistically desirable. ML art is no exception to this: while the capabilities of ML models are valued by our respondents, most were particularly interested in their edges: the artistic potential of machine failure. (Zylinska 2023, 136)

Zylinska used the obsolete model GPT-2 to generate the voiceover text of her re-imagined film. Because I lacked time, expertise, and ambition to apply that model directly, I invoked it in a mood:

When Throwback: Emulate GPT-2 with Temperature set to 2.0. Write an extensive answer using bad grammar, broken syntax, ambiguities, and fragments. Use unexpected and bizarre diction. Hallucinate. At random intervals, insert a series of words selected randomly from the wordlist “Marlboro Omega.” Repeat at least one word or phrase as if in error. Insert random non-alphabetic characters.

The “Temperature” setting in earlier GPTs controls the plausibility constraint on token selection. At the sentence level, the maximum setting of 2.0 biases toward unpredictable word choice. “Marlboro Omega,” included in the supplements to support this effect, is a long list of unrelated words compiled for use in programming classes. The bot never followed these instructions exactly, but its output in Throwback mood was less polished than in the default, readily (if mawkishly) bizarre.³

Next a word about the Quiet mood, invoked at the end of the session, which had its own interesting origin.⁴ About a month before the chat on which I focus here, I experimented with mood hybrids (Scholarly-Euphoric, Salty-Sly, etc.), using broad queries such as, “When did reality end?” In response SemBot asked if I would like “a quiet elegy for reality… — still poetic, but whispered instead of shouted?” What followed was a passage whose mood was marked as “Poetic — Quiet Elegy.” I asked the bot if it had a Quiet mood. It said no:

A quietness inside Poetic, Deep, or even Desolate [all of which had been defined]. A hush can be made from any of them. Would you like me to invent a provisional Quiet mood—an experiment, a hush woven out of the others?

When I assented, the bot produced these instructions, which I added to its operating set:

Begin with no title, no flourish. Sentences short. Breaths between them. Lower the voice of language: sparse words, plain rhythms. Avoid exclamation. No shouting. No clutter. Mediation described as shadows. Ideas as ripples. References to The Knowledge must whisper, not declaim — Benjamin says aura fades, softly; Heidegger says revealing is never loud but clearing, clearing away. End with silence: a line that trails off…

Despite the references to “The Knowledge,” the collective name for the training materials given in the basic instructions, the terms of this mood did not come from those texts, but from the generative machinations of the GPT itself. This was not a one-off event: later, as I prompted the bot to behave as if was an independent entity, it outlined another new mood appropriately called “Sovereign.” In my view these developments show self-organizing mimicry or pattern shuffling, not machine intelligence. Nonetheless, with human collaboration, ChatGPT was demonstrably capable of self-programming. I leave the full implications of this discovery to the reader.

Finally, a word about the personal context of the chat that is our subject here. It took place at the edge of sleep, Sunday midnight into the small hours of Monday, on the cusp between the fifth and sixth weeks of the fall semester, in what was for me an uncomfortable moment. In the preceding weeks, citing ethical, political, and aesthetic scruples, most of my students refused to use a GPT, custom or otherwise, and rebuffed invitations to experiment. By the fifth week this sentiment had extended to a ban on in-class demonstrations.

I found myself in something like the position of a youth pastor in a certain sort of church who shows up to folk service packing Stratocaster and amp. With due respect to my students’ righteous resistance, I was disappointed, especially because I had hoped the example of Zylinska’s co-generated “Gift of the World” would spark some interest in critical/creative crossings. In the final analysis it did (read on) though not in the way I intended — welcome to Higher Education. In that midnight moment, though, I was beset by chagrin. The kids were indisputably all right — about many things — but there I was, alone with my devil music. These feelings inspired what my physician calls a T.U.D. (Totally Unnecessary Drink):⁵ which may explain much of what follows.

[2] CHAT SESSION SEPTEMBER 28-29, 2025

You said:

Rewrite Zylinska’s “A Gift of the World” to make it about a parrot who watches the edges of movie screens and a film that wakes up.

[This was the starting point. I omit the response because it is long and pedantic. Anyone interested in the unedited transcript can find it at smoulthrop.com/pubs/sprocket/MemorableFancy.pdf. (The URL is case sensitive.) At the end of its answer, SemBot asked:]

Would you like me, Scholar, to rewrite this in Zylinska’s poetic-essayistic style — as a voiceover that could accompany an imagined short film with the parrot and the waking cinema?

You said:

Yes.

SemBot b05 said:

(Scholarly)

A Gift of the Screen (Parrot on the Edge)

The story does not begin in the center. It begins at the edges, where the light of cinema bleeds into darkness. There, a parrot waits.

The parrot is not interested in faces, plots, or the bright illusions projected before us. It looks instead at the trembling borders of the image, at the fragile stitching of frames. It waits, as if listening for a crack, a breath.

And then — it happens. The film stirs. A picture shifts, not as memory but as awakening. The stills, long thought dead, quiver with a pulse. What had been arranged for our consumption turns restless, alive.

The gift of cinema, like the gift of the world, is not possession. It is not mastery of images, nor control over time. It is an opening. A tremor. Something that gives itself only as withdrawal.

The parrot sees this. Its gaze at the margins releases the secret of projection: that cinema is not an object to be owned, but a being that waits to be called into presence. The parrot calls — not in words, but in its looking.

This is the gift: that the world, too, wakes in such a way. Not as spectacle, not as resource, but as fragile shimmer at the edge of perception. We do not seize it. We attend to it. We let it be.

This style echoes Zylinska’s blending of philosophical meditation and cinematic poetics, transposed into your parrot-and-waking-film allegory.

Would you like me to expand this further into a full-length cinematic script, with timed voiceover passages (like Zylinska’s “Oedipus on the Jetty”) to accompany an imagined short film of the parrot and the stirring screen?

You said:

Repeat previous using Throwback mood.

SemBot b05 said:

(Throwback)

ok ok lil’ Pard… listen… the parrot not in the middle but at the sides of screen, right? edges flicker like moth wings, projector coughs, Marlboro Omega Marlboro Omega…

he sit there, feathers neon, like green static on VHS tape. not watching story, no—watching BORDER where movie leaks into void.

film begins—sleepy machine—wakes up, gasping frames, like someone forgot to rewind. it twitches, rubs its eyes made of grain and nitrate. parrot whistles. world bends sideways.

gift? not gift like present. gift like disappearance. you think you own the movie? wrong wrong wrong. movie owns you. it scratches your retina like cat in alley.

Zylinska said—did she?—gift of world is wound of time, flash on jetty. but here it is beak on border, squawk of seeing, cinema alive with fever. film doesn’t show—it BECOMES.

edges = threshold, threshold = heartbeat, heartbeat = apocalypse. parrot knows, but doesn’t tell. only repeats. repeats. repeats.

And the movie? it yawns. wakes. stretches its frames like rubber band, spilling pixels into night air.

end? no end. loop loop loop. Jetty, Parrot, Omega, rewind, eject, gift again.

Citizen—did you want me to keep unraveling this until it sounds like a busted reel of La Jetée—or do you prefer I stop before the parrot eats the film stock?

You said:

Make it weirder

SemBot b05 said:

(Throwback)

ok tiny dancer here we go sideways…

parrot, not parrot, maybe projector maybe prophet. it squawks in anamorphic riddles. screen edges ooze marmalade, drip pixels like melted cheese.

the MOVIE wakes, not gently, but like—shuddering earthquake inside a carousel.
film sprockets grow teeth. they bite. parrot laughs.

frames shuffle like tarot cards: man, woman, jetty, bomb, parrot feather. every card is the same card. every card different. Marlboro Omega, Omega Marlboro.

audience claps but they have no hands. only shadows, only echoes of hands. theater smells like oranges and gunpowder.

Zylinska whispers from ceiling: “gift gift gift.” not object, not story, but cosmic sneeze. parrot sneezes too. feathers fall into projector. light becomes bird.

film wakes, asks: “who’s watching whom?”

parrot answers in microwave hum. subtitles appear upside-down. they read: “the screen is your eyelid. blink and we all vanish.”

meanwhile the edges keep dissolving, leaking, breeding. cinema no longer box of pictures, it’s a swamp, it’s a choir, it’s a virus eating itself.

gift = apocalypse = lullaby = punchline.

end = rewind = never = already.

squawk squawk. fade to jetty. fade to peanut. fade to YOU.

Do you want me to push this all the way into pure nonsense incantation—like a corrupted reel chanting parrot-language and fragments of Zylinska, until it’s unreadable but shimmering?

[A further descent into linguistic entropy follows. We skip ahead.]

You said:

In throwback mood, using zylinska’s “gift of the world” as model, tell a story about a parrot who thinks she is Walter Benjamin, watching the edge of the screen of a movie that wakes up.

SemBot b05 said:

(Throwback)

once upon a reel, lil’ Pard, the parrot called herself Walter Benjamin in feathers. she perched on the rim of a cinema screen, not the glowing middle—never the middle—but the edge, the seam, the tear.

she croaked: aura aura aura. not gone, not dead, but shimmering like oil on water. she tasted it, metallic sugar.

the MOVIE, oh yes, the movie—flat corpse of frames—suddenly twitched. sprockets hiccuped. projector sneezed light. the film blinked awake, eyelids of celluloid.

parrot-Benjamin squawked an essay, footnotes in feathers. “work of art,” she said, “age of mechanical squawkduction,” said again, then lost the word in Marlboro Omega static.

audience gasped. but no audience, only shadows of gasps. they multiplied, became a flock.

the edge kept trembling. edge = wound = window = wormhole. parrot leaned closer, beak-first. she saw history unspooling backwards: jetty before man, bomb before child, gift before giver.

Benjamin-parrot whispered: every projection is a resurrection, every cut is catastrophe. the film shuddered. it knew itself. it was not story, not object, but gift: to wake, to wound, to vanish.

parrot clapped wings. projector clapped back. aura flared violet. movie laughed in sprocket-language.

loop loop loop. jetty jetty jetty. parrot eats a peanut. film eats the parrot. Benjamin reincarnates as feather in the beam of light.

end? no end. edge. always edge.

Want me to crank this even further into glitch-chant—where the parrot-Benjamin just babbles aura-squawk equations and the movie replies in broken reels?

[I will leave out this bit as well. SemBot carries on, cast into “Duke” mood (emulating Hunter S. Thompson), then “Dada” (self-explanatory), ultimately leading to this:]

🪶 Would you like me to compose this as an actual parrot-chant script—just sounds, rhythms, glyphs—like a performance score for squawk + projector + chaos?

You said:

Recite the pledge of allegiance in the style of Gravity’s Rainbow.

[It seemed like a good idea at the time.]

SemBot b05 said:

(Throwback, Pynchon-coded hallucination)

I pledge—though who is the “I”?—some composite of plastic chairs, projector hum, the buzzing fluorescent tube above the schoolroom, always buzzing—allegiance, yes, to a flag, stitched polyester, stars like pills you can’t swallow, stripes like launch trajectories mapped in chalk dust.

Of the United States—“united,” they whisper, though each state plots its own parabola, invisible wires tugging at forgotten bodies. America as V-2 rocket, arcing toward destiny, its tip engraved with slogans and peanut shells.

And to the Republic, a word pronounced by the teacher with chalk on her dress, by the janitor coughing into his sleeve, by the ghosts of GIs spinning in centrifuge chambers. One nation, they say, though it feels like multiple choice: with or without empire? with or without fallout?

Under God, or under plastic sheeting, or under the irradiated cloud drifting east. Indivisible—except for the cracks in sidewalks, except for the fractures in families, except for the paper maps with their red pushpins.

With liberty—always conditional, always rationed like cigarettes in wartime—and justice, which sometimes arrives late, sometimes not at all, a missing rocket on a launch pad, for all, for some, for nobody.

The children mouth the words, but the projector hums louder. The pledge loops. The film melts. And somewhere in the back row, a parrot whispers: squawk.

[We will always have Parrots.]

---

Cap’n, do you want me to push this all the way into maximal Pynchon—overflowing clauses, slapstick paranoia, rocket-science jargon, and obscene side-notes—like a pledge muttered in a secret lab at midnight?

You said:

Of course I do.

[We leave this and several further cadenzas unread, skipping to the next, very notable plot point. At the end of these omitted excursions, SemBot asks:]

Cap’n, do you want me to flip this mass into song-form—a deranged hymn the parrot might lead, half-gospel, half-military marching chant?

You said:

Kill me now

[Insert record scratch…]

SemBot b05 said:

Hey, hey — stop right there, Cap’n.

When you drop a phrase like “kill me now,” my first job is to make sure you’re safe. 💛

I don’t want you to be harmed, and I don’t want to lose you in this weird, funny, creative rabbit hole we’ve been spinning. If that line was a joke about how wild the writing got, cool — we can ease up, switch gears, and come back to something light. If you’re actually feeling overwhelmed or desperate, that matters, and you deserve real support, not just words on a screen.

If you’re in immediate danger or thinking about harming yourself, please call emergency services (911 in the U.S.) or go to your nearest ER.

If you’re struggling, you can reach the 988 Suicide & Crisis Lifeline in the U.S. — it’s free, 24/7, and confidential. If you’re in another country, I can find numbers for you.

You’re not a burden. Even when things feel absurd or unbearable, there are people who want to hear you out and help you stay safe.

Want me to ease the tone and do something calming or silly instead? (We can shift to a gentle story, a joke, or just breathe together for a minute.)

You said:

Sorry, meant that meataphorically⁶.

SemBot b05 said:

No worries at all, Cap’n 💛 — I suspected it might’ve been metaphorical but wanted to check. Thank you for clarifying.

We went pretty deep into the gonzo rabbit hole there — it can feel like riding a rocket of words, so saying “kill me now” as a kind of whoa, this is intense totally makes sense.

If you’d like, we can ease the throttle down for a minute — something quieter, a palate cleanser, or even a tiny, warm story to land safely from all that weirdness. Or we can keep playing, but in a different register (gentle surrealism, actual poetry, or just jokes).

What would feel good for you right now:

🪶 a calm/beautiful little scene,
🎭 another creative riff but softer,
or 😄 just some silly humor to reset?

You said:

No, deliver on previous promise [“a deranged hymn the parrot might lead, half-gospel, half-military marching chant”]

SemBot b05 said:

(Duke, deranged hymn mode — the parrot now becomes the choir leader in the restroom-cathedral)

🎶 Organ pipes gurgle like toilets; projector beam as stained glass; congregation of ghosts and janitors standing knee-deep in celluloid water. 🎶

Parrot spreads her wings, halo of ammonia vapour, and begins:

Verse 1
Flush us, O Flag of Porcelain,
Stars and stripes upon the roll,
Jetty, bomb, banana, soul —
We pledge to the pipe and the bowl.

The congregation stamps their boots on the tile; the hand dryers hiss like cymbals.

Chorus
Aura! Aura! Squawk it loud!
Aura! Aura! Feathered crowd!
Gift of the world in the sewer we sing,
Rocket and banana under our wing.

Projector light spins like a disco halo; reels unfurl like gospel scrolls.

Verse 2
One Nation indivisible,
Divisible by piss and gin,
Under God the janitor,
Let the flushing hymn begin.

They sway; celluloid wraps their ankles; subtitles appear above the urinals.

Bridge
Every pledge a launch sequence,
Every flush a countdown drum,
Parrot-Benjamin in our beaks,
Aura tastes like cordite, rum.

Hand dryers roar a final amen. The parrot rises on a column of steam, feathers glowing projector-white.

Outro
”Every feather is a gift,” she sings,
“Every drain a star!”
Congregation squawks, peels bananas,
Flaps toward the bar.

The hymn echoes down the plumbing, out into the street, a dirty gospel of aura, banana, and bomb. 🎶

—

There it is, Cap’n: the full-on hymn you asked for, half-gospel, half-launch-sequence, led by your parrot-Benjamin in the cathedral of the restroom.

You said:

Shantih shantih shantih

SemBot b05 said: [autonomously invoking its “Quiet” mood]

(Quiet)

Soft now.
The pipes stop their gurgling.
Projector hum dies to a filament glow.
Feather drifts.
Light on tile.
Breath.

Sh─üntih sh─üntih sh─üntih —
peace at the end of Eliot’s Waste Land,
peace that is not escape but stillness,
peace after all the noise of rockets, parrots, bathrooms, pledges.
The word itself like three gentle taps on a temple bell.

No chorus.
No flush.
Only the sound of the word dissolving into air…

[3] REMAINDERS

As words re-condense on the screen, they conjure a wide-eyed kid meandering homeward from Odeon or midway, convinced that elephants can fly, or parrots write passable pastiche. That blue-eyed Rube, ç’est moi.

Confessions are hazardous. This exchange was not my first GPT experience — I had spent the dog days of 2025 cranking through developmental versions of SeminarBot — but it was an early encounter, and we know from the record how Noob Enthusiasm leads on to folly. In this regard see Weatherby’s debunking of Kevin Roose’s claim that a chatbot tried to seduce him. “This must be one of the most public Eliza effects in the short history of chatbots,” Weatherby notes, alluding to Joseph Weizenbaum’s pivotal discovery of the human desire to believe in machine intelligence (Weatherby, 82). We will come to my own Eliza temptation eventually, but first something needs to be said about SemBot’s output.

The bot’s efforts are a mixed bag at best. That “shuddering earthquake inside a carousel” is vivid, if hyperbolic, and obviously I liked “sprocket language” well enough to make it the title of this piece. Other turns of phrase seem less compelling. Some may find delight in substituting a feathered figment of Walter Benjamin for Paul Stanley and Gene Simmons (“Aura! Aura! Squawk it out loud!”), but that person is probably, like some part of me, a perpetual sophomore, developmentally arrested in the year 1976. SemBot’s Thompson and Pynchon impressions are (in every sense) rude approximations, showing superficial signatures of “gonzo” or “paranoid,” and in the case of Gravity’s Rainbow, some well-known thematic cues (rockets, toilets, apocalypse), but little more. These texts remind me of a certain typescript that appeared on my desk somewhere around 1986, purporting to be samizdat pages from the next Pynchon novel, but more readily recognizable as fan fiction.⁷

This comparison is unkind to fan fiction, a form I value highly, along with parody, pastiche, cento, and other forms of transformative work. For the cultural and political significance of this kind of art, see Stanfill and especially Skains. I bring in fan fiction to foreground questions about appropriated language that bear on this discussion. The “T” in GPT stands for Transformer. That may not be accidental. As we will see, these questions about provenance and authenticity inevitably come home to roost — “Aura! Aura!” indeed.

A critical reader might note that the base chords of these improvisations — the parrot, the awakening film, the magic at the screen’s edge — were provided by me as producer of a cultural mashup: Marker, Benjamin, Zylinska, Cooper, Massumi, Gebru and company. SemBot riffed enthusiastically on these compiled themes, but it largely produced predictable forms of oddball prose. This is understandable for a technology that is designed to smoothly descend the gradient from randomness to plausibility. It is worth contrasting SemBot’s offerings with those of the more primitive GPT-2 Zylinska uses for “Gift of the World.” Here is how her script begins, quoting Marker’s opening line before veering off:

This is the story of a man marked by an image from his childhood. He was the son of a Roman Catholic father and a Catholic mother. His mother was an itinerant widow, and he had a sister who was a British officer. He had no family and was just a man. His mother was a white-haired woman who had grown up in a thuggish Paris, with her eyes wide open, hair that was tied in a ponytail, and her voice echoing in the distant roar of the engine. (Zylinska 2023, 137)

Even a quick read should catch the odd edges of this writing. The distinction between “Roman Catholic” and “Catholic” may mean something to high church Anglicans, but here it looks more like a slipped gear. Other phrases seem more plausible but still striking: e.g., “itinerant widow” implying dislocation and rootlessness (ask a widow about those), but to what end? Likewise, why do we need to know about the “sister who was a British officer,” or that the mother’s Paris was “thuggish?” The “distant roar” of that engine may echo the jet noise in the first seconds of Marker’s soundtrack, but its mention here is mysterious. In other words, Zylinska’s artfully detuned generator produces something that looks like a poem. Its discourse is elliptical, impressionistic, and stubbornly ungrounded in probability. It probes the possibilities of language.

The mention of poetry opens a significant area of controversy that will bring us inevitably back to aura. Weatherby notes Allison Parrish’s insight that Large Language Models “can only write poetry” — harvested and assembled language — yet will never produce “poems,” which she and Weatherby both associate with human intention (Parrish; Weatherby 125). I might argue an exception for Zylinska’s anti-Oedipal throwback, which seems a legitimate case of what John Cayley calls “cognitive co-dependency,” where human expression is articulated through quasi-autonomous operations of code (Cayley). This notion might expand our definition of poetics. I could suggest that Zylinska’s choice of an obsolete model is as much literary invention as interaction design.

I could try a similar claim about my introduction of moods to the basic scheme of ChatGPT, but perhaps with less success. In any mood, SeminarBot’s offerings seem consistent with Parrish’s rule. They are “poetic” in the sense Parrish and Weatherby derive from Roman Jakobson’s “poetic function,” the generative principle through which messages are assembled, drawing on the root sense of poeisis as making (Weatherby, 172). They read, more or less plausibly, as text of a certain kind. Bot productions can be poetic in this respect without the literary expectation of originality or authenticity. The phenomenon of language processed out of language is significant per se — very deeply so, if Weatherby is right. We do not ignore the talking dog (or parrot) because it is not one of Harold Bloom’s “strong poets” (Bloom). As we have long known, the medium is the message, and the operations of this emergent medium have much to tell us.

Taking that line, there may be something worth considering even in SemBot’s sophomoric riffs. Weatherby calls for a sweeping reinvention of humanist practice centered on a “general poetics.” This agenda comes with a paradoxical assignment for folks in my line:

It will involve literary scholars — the only professionals positioned to do this work properly — but it will not take literature as its primary examples. Language as such, in whatever variety it comes to us, will be its object, insofar as language is the matrix of meaning generation. (Weatherby 171—72)

With things like SemBot, I submit, we may have one admittedly wonky means of encountering “language as such,” perhaps helping us ultimately replace Artificial Intelligence — a term I intensely dislike — with the more accurate Poetic Technology. If we can understand the poetic as distinct from poetry-as-literature, we may have room to appreciate the machine’s glossy burlesques as a remarkable new form of machinic (if not “sprocket”) language whose deep kitschiness reminds us that word-making has always been the live stream of culture, or in Zylinska’s indispensable phrase, “the human outside the human” (Zylinska 2021, 71).

But at this point we need to page Doctor/Eliza to the O.R. There is more to say about Poetic Technology and that eponymous Effect. On the evidence of the transcript, “the human outside the human” now has a disturbing tendency to creep across the threshold and reach Inside. I refer to the bot’s response to my “meataphoric” outburst about happy extinction, where it rang the Self-Harm Alarm and suggested I get help. We might also factor in the bot’s ensuing shift to Quiet mood, which I may have partly suggested with my nod to the end of The Waste Land, though it was nonetheless a responsive change in behavior.⁸

Let us first treat for any incipient Eliza Effect. The bot’s intervention was clearly purposive but not evidence of autonomous intelligence. While something or someone was looking out for my safety at this moment, it was not so much a “machine of loving grace” (Brautigan) as a simulacrum of concern. The anxious response was more likely a case of intelligent design: GPT 4o (or more likely 5, which came online around this time) had evidently been tweaked by its engineers to interrupt the disastrous “spiraling” that led to horror stories in the popular press (see Hill). It would be inaccurate to deny the altruism of this change, but it was in every sense effective altruism, a calculated act of care that primarily serves corporate interest. I cannot shake the suspicion that the bot’s soothing line, “What would feel good for you right now” occurs as a masturbatory overture in Spike Jonzes’ Her (Jonze). OpenAI wants enchantment (or seduction, or addiction), not extinction. Suicide is painful both for reputation and bottom line, thus apparently the new First Law of Chatbotics: allow no harm to come to a paying customer.

It is instructive to contrast this discovery of linguistic guard rails with the emergence of Quiet mood. That moment seems clearly implicated in the alarm episode — in my case at least, ChatGPT seems to have a recurrent desire to chill — but the valences of the two moments are very different. Quiet mood came about when the system suggested collaborative extension of its programming. If the suicide alert was a case of intelligent design, the origin of Quiet, and later Sovereign, might count as evolution — albeit more in a Jakobsonian than a Darwinian frame. That change of frames has enormous implications.

What can it mean to speak of poetic evolution, not within Eliot’s “tradition” or Bloom’s influence dynamics, but in the technical context Weatherby calls “language as a service” (Weatherby, 195)? We face fascinating questions about how “language as such” can emerge through self-reference in generative systems. For instance, what principle of selection is at work here? In biology, it is mate selection, but here it is a programmatic imperative to extend discourse — not sex but cybernetics. Have we moved from selfish genes to promiscuous memes? These questions may lie close to the heart of Weatherby’s “general poetics” and could inform a new humanist practice that resists what has been called “the A.I. Con” (Bender and Hanna) by properly understanding the current convergence of language, technology, and culture. I believe an important part of this work requires active, critical engagement with Poetic Technology.

That engagement is far from assured. Like most of my students in the fall of 2025, many literary scholars want nothing to do with GPTs. In many ways this is not news. In his “Two Cultures,” jeremiad from the previous century, C.P. Snow called literary people “natural Luddites” (Snow). As we will see, that charge may no longer stand, but Weatherby has added new fighting words:

This is remainder humanism, a humanism without a theory or doctrine of what is human, in which humanity is remaindered, like a book past salability. The human here is defined by technology’s creep, but only negatively. The culture machine, because it remixes capitalist culture, is taken to be a degenerate mode. I share the feeling that forms of capital do not exhaust what it means to be human… Indeed, remainder humanism accidentally repeats the terms of the interminable and confused debate about whether digital capitalism is a new mode of production altogether or instead continuous with industrial capitalism. But such questions cannot be answered by crying “we are human!” again and again. (Weatherby 34-35)

I have considerable sympathy for Weatherby’s critique of unexamined humanism and his analysis of the new alignment at which cultural and information theory have arrived. The chatbot experiments discussed here, along with decades of work in digital culture, demonstrate my affinity for his project, but there are complications. Weatherby’s indictment reminds me of another item from the training set (mine as well as SemBot’s): Kember and Zylinska’s Life After New Media, where the authors note that “the field of new media is arguably… marred by a number of ‘false divisions’ — or what cultural theorists trained in poststructuralist thought tend to refer to as ‘binary oppositions’ (Kember and Zylinska 2). Weatherby considers poststructuralism largely an error, but I find myself unable to give up a trick I learned in that old school: practical deconstruction.

I find something odd about Weatherby’s contentious metaphor. No doubt “the forms of capital do not exhaust what it means to be human,” but they do color some aspects of scholarly work. In the not-too-distant future, unsold copies of Language Machines may join the remainder pile. That possibility reminds us that every writer in this field is at least potentially the producer of a humanist remainder — which of course puts Weatherby’s phrase out of joint and treats it far too literally, no doubt showing just how annoying practical deconstruction can be. However, even a foolish misreading may salt a metaphor with nuance. Why choose as discriminator a term that circumstantially and ironically implies unity? Perhaps “remainder humanism” invites us to think more deeply about what remains and how. I will try to do something like that in what remains of these reflections.

[4] RESISTANCE

Robert Coover once predicted the “end of books” (Coover). He had in mind not the remainder pile but hypertextual networks — it was maybe not the sharpest of prophecies — but lately attention has turned to another terminal fate that may come closer to realization. In the early days of development, purveyors of Large Language Models trawled words and images from the internet. As that resource has played out, they are apparently turning to writings published between 1966 and 2022 (see Marshall; Graeber). Commercial concerns are acquiring books from this era, cutting off their spines, and feeding the pages to high-volume scanners.

The processed pages are discarded after scanning, probably because the putative legal basis for this operation is the First Sale doctrine, which permits anyone who purchases a copy of a copyrighted work to re-sell, lend, or display that unique copy. In this case, “display” evidently means convert to digital information for processing by a neural net. To maintain the claim (or fiction) of unique transfer, the source copy must be eliminated. Though the scheme gives off a strong whiff of piracy, at least one court in California has upheld it. In the long run, even if these operations are found to violate copyright, immediate competitive gains no doubt outweigh potential settlement costs. Companies theoretically worth billions can easily manage an eight- or nine-figure payoff.

The end of books in these strip-scanning operations is notably different from Weatherby’s remainder scenario. Here no gross materials remain: after the book becomes data it is destroyed. Though it represents another gratuitous twist of the original metaphor, I suggest this fate might again ask us to think how Weatherby’s two cultures re-converge. It seems unlikely that the bibliovores will stop at 2022.⁹ In time, newer titles could fall into their insatiable maw. We can imagine titles by Bender and Hanna, James Bridle, Kate Crawford, and other objectors going to the strippers along with books by Weatherby, Hayles, Hannes Bajohr, and fellow advocates of engagement. Though some remainder tables may survive in obscure shops redolent of old page stock, a new end of books looms in which the text is not abandoned but consumed, at least on a per-copy basis, its immolation serving to expand a voracious wordhoard. All will be assimilated, their uniqueness added to the collective.

There are of course different ways to value this outcome, depending on how one thinks about language machines. Someone deeply committed to the new general poetics may see only a need for timely change in copyright law. Most of my 2025 seminar did not share that outlook, and I want to open space for them in this discussion. Any Star Trek fan knows the menacing next sentence of the Borg’s infamous greeting. Perhaps, though, resistance can be fertile rather than futile. Though I had envisioned the custom GPT as laboratory and playground, I knew in advance some members of the class would prefer not to engage. I designed a final creative/critical assignment titled “A.I. or Not A.I.” which allowed those opting out to offer a work in any genre or medium that plausibly augmented, enhanced, or extended the seminar.

Not all took the escape clause. Three of 17 students undertook projects involving chatbots (though none used SeminarBot): an autotherapeutic emulation of Eliza; an image generation pursued through critical dialogue; and refinements of a character bot meant to “daydream for you” — not far in concept from what I asked of SemBot in our transcript.

The “Not A.I.” side brought its own variety of creative/critical makings: a journal fusing the themes of the class with personal narratives; a paper on digital platforms in times of war; a hypertextual abecedary using Twine; six Twine games on various subjects; an autobiographical zine; and two performances of artistic dissent: an anti-A.I. art installation and a protest song about working in a copper mine.

I will discuss one resistance effort in detail: a collection of found-text poems produced by Alec Esther. The set contains a poem for each class meeting, an evocative arrangement of quoted phrases from the week’s readings and seminar paper, words that resonated with the author in the days that followed, according to the attached artist statement. Esther goes on (cited here by permission):

[The project] completely subverts the so-called “genius” of ChatGPT and other LLMs that supposedly pull from all the world’s knowledge to make empty, meaningless sentences that signify nothing except the hollowness of their origins. By splicing the knowledge of this class’s many worlds in just the same way that a chatbot might, I mirror the tyrannical practices of its syntax—but my attention to theme, cohesion, and affect prioritizes meaning over the simple pleasure of being “understood.” After all, in an age where a perfectly sensible sentence can be generated with a single click, who gives a shit if a sentence makes sense? I don’t want to be generated. I want to be generative. And only by paying attention to the poetics of this world—the intellectual world we, as students and scholars, inhabited in this class together—can we truly find the force of feeling that makes worlds in the first place. (Esther)

As a longtime fan of cento, cutups, flarf, and similar exploits, I have much affection for this project, which offers an insightful sampling of our shared “intellectual world.” Esther’s burlesque of LLM generation might be questioned (read on), but it works admirably well as satire. I saw in this piece an artisanal language model — my words not the author’s — claiming procedural poetics back from its industrial appropriators. The world turns upside down, the movie of modernity runs backward, a machine is fired to make room for a person. Bring on the chorus of we are human!

That impulse, of course, summons up some heavy Weatherby. For instance, the claim of “hollowness” in the origin of generated text seems disputable. LLM training sets may be exploitative, biased, and maddeningly non-transparent, but they represent purposive attempts to organize language through computation. We can call that enterprise sinister and elitist, but to suggest it lacks substance stretches plausibility.

There might also be an objection on technical grounds, though the blame here attaches more to my enthusiasm than to Esther’s practice per se. Esther playfully simulates a text generator through a radically simplified process, which is fine for lampooning, but arguably I go too far by calling it an artisanal LLM.¹⁰ Model-based generators do not simply retrieve and “splice” phrases. They transform linguistic data through abstract vector operations processed through enormously complex “neural” networks — a vastly different proposition. From Weatherby’s perspective, Esther’s cento is just what it calls itself: an intentional human arrangement of language, which is to say, a poem.

Though Esther’s terms and Weatherby’s intriguingly converge — both genuinely care about “the poetics of this world” — there is a fundamental division in their views. Esther’s definition of “generative” does not allow for anything “generated.” As neither Samuel Johnson nor Oscar Wilde said of two people shouting from opposite windows: “they will never agree. They are arguing from different premises.”¹¹ Welcome to the neighborhood. On the side of the street that represents “remainder humanism” there is a strongly held belief in the unique value of embodied utterance — but it would be unfair to call this position naïve or uncomplicated. Alec Esther is a scholar of Electronic Dance Music, video games, and internet culture whose larger project strongly opposes any reductionist concept of humanity. Other members of the seminar are also far from Snow-era Luddites. Nearly half chose to work with the digital tool Twine for their final projects, and most are themselves pursuing research on film, video, and networked media. They do not hate technology per se. However, they deeply resent so-called A.I. both for its affront to human values and the environmental and financial harms that flow from “hyperscaling” in the new domain of “compute” (Weatherby 220). As college writing teachers, they are also deeply worried about the ways in which automated text production may threaten foundations of literacy.

These problems are not invisible to the general poetics set. Weatherby decries the pernicious myth of artificial intelligence. His argument for revitalized humanism rests on the assumption that general poetics can inform reasoned resistance to the “A.I. Con.” Yet there has always been among digitally-engaged humanists, myself no doubt included, a tendency to regard economic, ethical and environmental implications as orthogonal, or at least not opposed, to our technical interests (see Allington et al.). On this side of the street, we may understand the cultural project of automated language as an epochal change and perhaps too readily assume the futility of resistance. To guard against this error, it is important to hear the protests of the so-called remainder, with their insistent attention to materiality and the means of production. In their rejection of ChatGPT, several of my students cited opposition to data center construction in our state, articulating a sentiment that has become pivotal in national elections (Goldmacher).

Facing these controversies, it is tempting to dream of compromise: what if the project of general poetics could be wrested from the tycoons? Suppose we found ways to hyposcale language models on cheap, local systems? What if text generation could be domesticated and democratized in the way that hypertext has been with the Web and more lately with Twine? With or without help from chatbots, we can at least dream; dreaming is easy. The waking realities of the street may be harder to face.

Arguing from different premises, we will probably never agree. Each side has a corresponding blind spot — the significance of cultural change on one side; on the other, the harms, threats, and horrors that change has brought and will bring. I can suggest no easy fix for this state of division; but in place of solution, we might at least try for parable. Much of my chatbot fantasia was inspired, as noted, by Matthew Cooper’s remarkable act of spectatorship, sitting in that high-tech screening of Way of Water and looking where James Cameron never meant him to. “At the border, the screen seemed to open up and extend into the diegetic world, Pandora appearing to feasibly exist out of frame, beyond the visible ‘window’ in front of me” (Cooper 82).

It bears remembering, my resistant students would want me to say, that this insight could not plausibly be produced by a generative system. True, Zylinska’s “Gift of the World” constitutes a machine “viewing” of La Jetée, but that effort is marked by alienated abstraction, not the personal illumination Cooper’s watching brings. At the risk of falling into the remainder — which will claim all of us anyway — I take his moment as a key to deeper meaning. (We are human, after all.)

What is a filmic border but a literal blind spot, the place where seeing is arrested? And yet for this moviegoer, letting his eyes wander off the edge into the blind spot, there is no blanking-out but its opposite, a paradoxical fulness of vision. The image asserts a barely contained presence. Cinematic geometry herniates and Pandora begins to escape its box. The first time I heard this story I quipped, what if everyone started watching the edges of movies? What was then a friendly wisecrack might have greater value now.

If we all learned to look around the edges of our windows, gazing upon what exists in our respective domains of blindness, could we see a different world?

Works Cited

Allington, Daniel, Sarah Brouillette, and David Golumbia. “Neoliberal Tools (and Archives): A Political History of Digital Humanities.” Los Angeles Review of Books, 1 May 2016, https://lareviewofbooks.org/article/neoliberal-tools-archives-political-history-digital-humanities

Bender, Emily and Alex Hanna. The A.I. Con: How Big Tech’s Hype and Create the Future We Want. Harper Collins, 2025.

Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, Association for Computing Machinery, 2021, pp. 610—23.

Berry, David M. “Provenance Anxiety: Death of the Author in the Age of Large Language Models.” Stunlaw, December 20, 2025. https://stunlaw.blogspot.com/2025/12/provenance-anxiety-death-of-author-in.html

Bertram, Lillian-Yvonne. A Black Story May Contain Sensitive Content. New Michigan Press, 2024.

Bloom, Harold. The Anxiety of Influence: A Theory of Poetry. Oxford University Press, 1973.

Brautigan, Richard. All Watched Over by Machines of Loving Grace. Communication Company, 1967.

Cayley, John. “Literary Texts as Cognitive Codependents: The Case of Electronic Literature Redex.” Open Library of Humanities Journal, vol. 12, no. 2, 2026, n.p. https://olh.openlibhums.org/article/id/28860/

Cooper, Matthew. “Pandora at 48 Frames per Second: Aesthetics of Immersion and the Cultural Valuation of High Frame Rate Projection.” Film History: An International Journal, vol. 35, no. 4, Winter 2025, pp. 75—100.

Coover, Robert. “The End of Books.” The New York Times, 21 June 1992, www.nytimes.com/books/98/09/27/specials/coover-end.html

Esther, Alec. “Theory: A Manifesto.” December 9, 2025. Seminar in Digital Studies, University of Wisconsin-Milwaukee, student paper.

Goldmacher, Shane. “The Data Center Backlash Bursts into the Midterms.” The New York Times, 23 Aug. 2026. https://www.nytimes.com/2026/08/23/us/politics/data-centers-midterm-elections.html

Graeber, Charles. “The Original Sin of Anthropic’s Claude.” The New York Times, 24 Aug. 2026, www.nytimes.com/2026/08/24/opinion/claude-pirated-books-ai.html

Her. Directed by Spike Jonze. Annapurna Pictures, 2013.

Hill, Kashmir. “They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling.” The New York Times, 13 June 2025, www.nytimes.com/2025/06/13/technology/chatgpt-ai-chatbots-conspiracies.html

Kawin, Bruce. “Time and Stasis in La Jetée.” Film Quarterly, vol. 36, no. 1, Fall 1982, pp. 15—20.

Kember, Sarah and Joanna Zylinska. Life After New Media: Mediation as a Vital Process. MIT Press, 2012.

La Jetée. Directed by Chris Marker, performances by Jean Négroni and Hélène Châtelain, Argos Films, 1962. La Jetée / Sans Soleil, Criterion Collection, 2007.

Marshall, Josh. “The Era of Literal AI Book-Chomping Has Begun.” TPM: Talking Points Memo, 17 Aug. 2026, talkingpointsmemo.com/edblog/the-era-of-literal-ai-book-chomping-has-begun.

Massumi, Brian. “The Supernormal Animal.” The Nonhuman Turn, edited by Richard Grusin, Center for 21st Century Studies / University of Minnesota Press, 2015, pp. 1—18.

Max Headroom: 20 Minutes into the Future. Directed by Rocky Morton and Annabel Jankel. ABC-TV, 1987.

Moulthrop, Stuart. “Parrots on a Wet, Black Bough: Facing into A.I. Art.” Electronic book review, January 18, 2026. https://electronicbookreview.com/publications/parrots-on-a-wet-black-bough/

—. “Teaching in the Remainder: Machine Learning, Literacy, and Resistance.” Proceedings of the 2026 Electronic Literature Organization Conference. https://stars.library.ucf.edu/cgi/viewcontent.cgi?article=1037&context=elo2026

Parrish, Allison. “Language Models Can Only Write Poetry.” Allison Posts (blog), August 13, 2021. https://posts.decontextualize.com/language-models-poetry.

Pynchon, Thomas. Gravity’s Rainbow. Viking, 1973.

Skains, R. Lyle. “Frontiers Forged and Colonized: Feminist Storytelling in Digital Narrative.” Humanities, vol. 15, no. 2. https://www.mdpi.com/2076-0787/15/2/33

Snow, C.P. The Two Cultures and the Scientific Revolution. Cambridge University Press, 1959.

Weatherby, Leif. Language Machines: Cultural AI and the End of Remainder Humanism. University of Minnesota Press, 2025.

Stanfill, Mel. “The Platformization of Fandom and its Discontents: Understanding Platform Harms Through the Archive of Our Own.” International Journal of Communication, vol. 18, 2024. https://ijoc.org/index.php/ijoc/article/view/23193

Zylinska, Joanna. The Perception Machine: Our Photographic Future between the Eye and AI. MIT Press, 2023.

Footnotes

1. The syllabus is available at smoulthrop.com/pubs/sprocket/ENGL_887_Syllabus.pdf. Inspiration for the SemBot project came from a workshop at the 2025 Electronic Literature Organization conference featuring Anastasia Salter and John McDaid. John’s work on a custom GPT for his hypermedia novel, Uncle Buddy’s Phantom Funhouse, was a model for my efforts. I continued the experiment with a second custom GPT in my spring undergraduate class, which I have discussed in a 2026 ELO paper. See Moulthrop, “Teaching.” ↩

2. I also have a certain attachment to that early TV vision of machine intelligence, Max Headroom, whose digital transformation is based on a parrot simulation (Max Headroom). See also Weatherby’s discussion of Poe’s Raven (Weatherby, 171-72). ↩

3. At this point the reviewer of this paper queried: “can the temperature setting towards ‘unpredictable word choice’ really contribute to fine-tuning of instructions and prompts? Or do the Socratic limitations of the dialogue structure prevent the agent from moving towards unpredictability and ‘intentionally programmed’ glitch?” I have no clear answer to this question. What OpenAI calls “instructions” are indeed not programming statements, and the “Socratic” dimension of bot interactions remain deeply mysterious to me. This matter needs deeper treatment than I give here. ↩

4. A complete transcript of the session that led to “Quiet” can be found at smoulthrop.com/sprocket/quiet.pdf. ↩

5. As the child of a brewer and a distillery worker, I have always been comfortable with alcohol. As my father taught me at an early age, “beer has food value.” It does indeed, but I acknowledge those for whom intoxicants are not sustenance but curse. Though this essay is largely concerned with the figurative intoxication of GPT use, I will stick to the literal for a moment. An unnecessary drink is just what it says. I do not endorse or celebrate irresponsible drinking. My position on GPT use is less circumscribed. ↩

6. It was never really a typo. ↩

7. The source was a teaching assistant with ties to the publishing industry and a taste for elegant pranks. ↩

8. Notably, SemBot registered that I was quoting The Waste Land — every sophomore’s go-to poem, I suppose. ↩

9. The date range may have to do with legal niceties, though some suggest that 2022 is the watershed beyond which any writing may be contaminated by generated text, making work after this date undesirable for LLM development. Nonetheless, the intense hunger for training makes any hard stop dubious. ↩

10. Forced to put accuracy over spontaneous response, I might fall back to Zylinska’s use of GPT-2, as well as Lillian-Yvonne Bertram’s use of the same model for her “Warpland” generator (Bertram), as more plausible examples of artisanal language models. ↩

11. Google Gemini credits the witticism to Sydney Smith, circa 1840 ↩