Opus Ipsum, the work itself

On PTFO, AI Writing, art, and feeling

PTFO:Stonetop is on a brief break between Session 16 and Session 17 — Anwen, Padrig, Vahid and Mado will return on June 8th. But as usual during hiatus, I still can’t quite help myself from banging on the keyboard a bit. Today I want to, perhaps unwisely, lie upon a bed of nails and talk about AI-generated fiction—the discourse surrounding it and my personal use of it on PTFO.

You may have noticed that this piece did not hit your email inbox — it’s only published here on Substack dot com. I know AI/LLM writing is not everyone’s favorite subject, and since people signed up for solo RPG actual plays and serial fiction, I decided not to shove it in folks’ faces.

Prizeworthy LLMs

Another week, another LLM-writing kerfuffle. One of the more recent concerns the Commonwealth Short Story Prize, which gave a regional award to a story called "The Serpent in the Grove" by Jamir Nazir. One of the judges, Sharma Taylor, described his prose as "polished and confident, with a melodic voice that lingers long after the final line." Within days, someone ran the story through Pangram, an AI-detection tool that claims 99% accuracy, got back a 100% rating for AI-generated text, and posted their results to X.

This led far, far more people to read the story than otherwise would—the Commonwealth Prize is prestigious, but that doesn’t mean it’s highly visible, compared to your average TikTok POV video. Readers quickly discovered lines like this:

She had the kind of walking that made benches become men.

And this:

The girl smiled like sunrise over a sink.

And this:

Doing is a treacherous bridge: you step on and it carries you to a side you didn’t plan to reach.

A Substacker named BDM, writing about the bridge line: “It does read as if written by somebody who doesn't really know what a bridge is or even what sides are.” Trinidadian author Kevin Jared Hosein, himself a former Commonwealth Prize overall winner, called the prize "dead" and said parts of Nazir's work "were purely nonsensical, and seemed like a product of AI hallucination."

And it isn't only him. Pangram flagged two more of the five regional winners — John Edward DeMicoli's The Bastion's Shadow (Malta) as fully AI-generated, and Sharon Aruparayil's Mehendi Nights (India) as partially AI-generated. The DeMicoli has lines like:

Her grandfather used to say that limestone remembered, but only because someone first pressed a memory into it. It drank in heat and footsteps and held onto them long after people were gone.

This metaphor is not quite so strained as Nazir’s work, but it does have that characteristic density of literary prose that AIs love to emulate, and that sends me running for the genre fiction aisle. People had a grand old time pillorying the alleged LLM prose, but I have read prose I disliked just as much written with hands of pure human meat.

One of my favorite takes on the LLM-hate-party came from writer Justin Murphy on X:

What if Jamir Nazir is just a sincere, passionate writer using technology to explore new approaches to literature, and Granta is now the most daring, avant-garde journal of contemporary writing? This emerging collective deference to Pangram among digitally cosmopolitan readers strikes me as more philistine than an Indian businessman trying to write poetry or tell stories with the help of his computer. Who here is really most guilty of outsourcing their taste to the machine?

Indeed. Once a text is determined to be created by AI, there is a subset of readers who cannot, or will not, engage with it on the same level as a purely human-authored text, regardless of its specific qualities.

I suspect some PTFO readers feel this way. If I can try to state this position as fairly as I can, it runs something like this: Writing, or any art form, is fundamentally a person reaching out to another person. It carries feeling rendered into language. When you discover the person was aided by a clever machine, the reaching out you thought you experienced never happened. And since that reaching out never happened, something written by or with an LLM can’t be art.

Now, Plato thought art was imitation, dangerous because it lied. Aristotle thought it was catharsis, valuable because it purged and clarified. Kant thought it was disinterested satisfaction in the beautiful. Tolstoy thought it was the transmission of feeling from one person to another. Oscar Wilde thought it was useless, and that was the point. John Dewey thought it was a heightened experience. The debate has ebbed and flowed for 2,500 years.

But today, in 2026, the presence of AI-generated art and writing has filled a lot of people with a powerful certainty about what art is, and more importantly, what it is not.

Everyone listen up, we’ve figured out this art thing

In the long debate about the nature of art, I’m a pluralist: I think the 2,500-year debate is like the parable of the blind men and the elephant — I have touched the elephant, and it is very like a tree, like a wall, like a spear, like a fan.

And I don’t have to commit, because I’m not really in the business of telling you what art is fundamentally about. I’m more in the heroes and peril and high adventure trade. But let’s say that the position above — the idea that we know what art is, and it’s a connection between two humans, and some amount of LLM spoils the brew — is capital-t True. If that’s the case, how does one detect that sufficiently human-to-human connection when beholding an artistic text?

Well, you just have to read it and see if you feel anything. You certainly don’t need to run the work through Pangram—your own imaginative and emotional equipment does the job just fine. The play’s the thing, not the players. All you need to do is look at the work itself, opus ipsum, and ask if it moves you. If it moves you, there must’ve been a human connection, and if it didn’t, it was a miss — maybe because the connection was ruined by too many em dashes and Not X But Ys, or maybe because their ideas just didn’t quite reach out and touch you. Many such cases.

A tour of the sausage factory

I’ve alluded to this before, but never stated it plainly: I use LLMs in the writing of our PTFO stories. I use them for research. I use them for editing. And I use them for drafting prose.

I think PTFO:Stonetop is a labor of love. I love the fantasy genre, I love TTRPGs and GMing. And I think that love comes through in the work, but the real test is whether readers feel something when they read it. Fortunately, I have some bona fides right here:

Proof that SGH has a heart.

This message made my week. And it’s true — I did get a little choked up writing that part of this episode. But my process for writing it was deeply enmeshed with my frequent writing companion, Claude Opus 4.6.

I’m going to pull the curtain back on my writing process and show exactly how PTFO:Stonetop’s prose comes together: The conversation I had with Claude, what it produced, and how I shaped it into the published piece.

Keep in mind, this is not a fresh, off-the-rack Claude chat—this particular model has been given detailed instructions on how to work with me. For the first 3 years of this project—hundreds of thousands of words—all the prose was written by me alone, with my very own meat hands. I hadn’t yet learned to use LLMs and the models were still in their infancy. This particular Claude project has access to about 75k words of PTFO prose, and can search all of our conversations going back to Session 10 or so. Without this baseline, I very much doubt I would get satisfying results.

Before I begin writing, I expand my notes into a beat sheet through a conversation with Claude, and armed with all that context, we write the fiction: Claude drafts, I redraft, Claude gives feedback, I redraft, and so on until we arrive at a published version.

We’ll start with an example from Session 16.1: The first scene, where Padrig is keeping watch, waiting for Mado and Vahid to return from their perilous delve into the Barrow of the Thrice-Betrayer. I had played through the scene and taken some rough notes, and then worked through a beat-by-beat breakdown through a conversation with Claude. Then, when I was ready to start writing, I prompted Claude like this:

And now let’s go through that scene, comparing the Claude 1st draft and the final published version side-by-side.

Claude:

The wind comes from the southeast, cold and sharp for mid-spring, carrying a bite that has no business on a night this late in the season. Padrig draws his cloak tighter and shifts his weight on the rough stone of the Old Wall.

From up here, the village is a scatter of low, dark shapes against the hillside, hearth-smoke rising thin and silver in the starlight. The nearest fields are dark furrows of turned earth, the barley shoots just visible as a pale green haze in the last of the twilight. Beyond, to the south and east, the rolling grasslands stretch into blackness, and on the horizon, a line of cloud blots out the stars.

The Wall itself is a ruin beneath his boots. The Maker-scale stones are still sound — they’ll be sound when the last man in Stonetop is dust — but the courses above are gapped and crumbling, the mortar long since eaten away by frost and rain. We’ll need to get to work on this. He runs a hand along the rough stone. Good bones, but it won’t stop a warband as it stands. Marged’s folk could quarry from the collapsed sections, use the rubble to shore up the gaps. He files the thought away for the morning.

Published:

The wind comes from the southeast, cold and sharp for mid-spring. Padrig draws his wolf-pelt cloak tighter and shifts his weight on the rough stone of the Old Wall.

The cyclopean stones beneath him are tall and sound—they’ll be sound long after Pad’s bones are dust—but the muster of Stonetop could never hold their massive perimeter. The inner ringwall encircling the steading was barely waist-high. We’ll need to get to work on this. He runs a hand along the rough stone. Marged’s folk and the miners from the Delve could quarry from here, use the rubble to build up the inner wall. If there’s enough time.

From up here, the village is a scatter of low, dark shapes atop the hillside, hearth-smoke rising thin and silver in the starlight. The nearest fields are dark furrows of turned earth, the young barley just visible as a pale green haze in the last of the twilight. Beyond, to the south and east, the rolling grasslands stretch into blackness, and on the horizon, a line of dark clouds blots out the stars.

Here, the beats have been reorganized (I wanted to do Pad’s interiority about the work to be done first as a reminder, to ground the beginning of the session), and the specific tactical thought has been rewritten to correct it — Claude often forgets Stonetop’s double-wall layout, and other setting details unless specifically prompted with them. Also, I love the word ‘cyclopean.’

Claude:

Ozbeg leans against the wall a few paces off, his shortbow across his knees, humming something tuneless. The old Peaksman has been in fine spirits since Anwen delivered Parvati’s coin — he wears it on a cord around his neck now, tucked beneath his leathers where it rests against his chest. Padrig has caught him touching it when he thinks no one is looking, the way a young man touches a lock of his sweetheart’s hair.

He has also caught Ozbeg making a fool of himself over Marged’s widowed sister at the well, and at the Public House, and twice at the granary where the woman has no earthly business being. Ozbeg has never been one to think too far ahead in matters of the heart. Parvati’s candle burning at his breast, and another woman’s laugh in his ears, and the old mountain bandit sees no trouble in it at all.

Published:

Ozbeg sits atop the wall a few paces off, his recurve across his knees, humming something tuneless. The old Peaksman has been in fine spirits since Anwen delivered Parvati’s coin — he wears it on a cord around his neck now, tucked beneath his leathers where it rests against his chest. Padrig has caught him touching it when he thinks no one is looking, the way a young man touches a lock of his sweetheart’s hair.

He has also caught Ozbeg making eyes with Marged’s widowed sister at the well, and at the Public House, and twice when he was on duty at the granary, where the woman had no earthly business being. Ozbeg has never been one to think too far ahead in matters of the heart.

Here, I cut some things that I felt were too flowery for my voice (the bit about Ozbeg with a candle at his heart, which my wife quite liked!)

Claude:

“Quiet night,” Ozbeg offers.

“Aye.” Padrig’s eyes sweep the southeastern horizon. Somewhere out there, Vahid and Mado are flying through this same cold wind, bearing Stormcatcher’s Crown back to Stonetop. Or they are dead in some sorcerer-king’s tomb, and the crown is lost, and everything they’ve fought for since the Delve is for nothing. Anwen told him they’d return tonight, if all went well. If all went well. He’s been a soldier too long to put much stock in those words.

“You’ll wear a groove in that stone, the way you keep looking south,” Ozbeg says.

“Go and check the eastern watchtower. I want to know if Pryder’s lot are awake.”

Ozbeg grunts and levers himself up, slinging his bow. “As you say, captain.” He ambles off along the wall, and Padrig is alone with the wind and the dark.

Published:

“Quiet night,” Ozbeg offers.

“Aye.”

“Cold, too.” He puts his hands inside his patchy yakhide cloak.

“Aye.” Padrig’s eyes sweep the southeastern horizon. Somewhere out there, Vahid and Mado should have this same cold wind at their backs, bearing Stormcatcher’s Crown back to Stonetop. Or they are dead in the sorcerer-king’s tomb, and the narrow gap through which their hopes shine is drawing closed. Anwen told him they’d return tonight if fate is with them. If fate is with them. Has fate been with us? When we failed at the Delve? When the Sun-Spear’s camp burned? We live, at least, we three who swore the oath over our hearth.

“You’ll wear a groove in that stone, the way you keep pacing,” Ozbeg says.

“I’m on watch, old hawk, you might be familiar with the notion,” he says quietly. “If you’re cold, get your blood moving and check the eastern watchtower. I want to know if Talfryn and Andras are awake.”

Ozbeg grunts. “If Vahid returns, we’ll see him from here. The whole gods-damned village will.” Oz has heard tell of what Vahid can do. Stormcatcher, some of the emigres still call him.

“As you wish. I’ll go,” Padrig moves to climb down from his perch atop the wall.

The most substantial edits in the scene. Claude’s Ozbeg/Padrig dialogue was functional, but they could have been any two soldiers on any wall. I extended their exchange to load in three things the published scene needed: Pad’s interior reckoning with the serial’s recent failures (Has fate been with us?), the oath that binds the three main characters together, and Ozbeg’s nod to what Vahid has become. This makes it part of the larger serial, which Claude often fails to do.

And so the episode gets written. I think I’m adding value here, but Claude laid down a strong foundation to work from after I gave it the scene’s beats.

That was a relatively edit-heavy passage. The passage from Session 16.6 the above reader DMed me about, the binding ceremony, was not. I have to confess, before I compiled the quotes from this piece, I had forgotten just how much of the prose was Claude’s first draft, but truth is good for the soul. I do remember getting misty as I compiled, edited, and re-read the work, though, so if I’m bamboozling people with my LLM-aided writing, my first mark is myself.

Let’s get into the blood-and-guts of how that scene got written (mostly by Claude). Here’s the prompt that gets us started—after a bunch of structural discussion about the episode:

As you can see, the plotting is all mine—rarely do I need Claude’s help in deciding what should actually happen next. In response, it gave me some structural feedback, asked a few follow-up questions, and then drafted.

Claude is at its strongest when drafting more zoomed-out montages, and I think the draft it produced is some of the best work I’ve seen it do, so you will note that my changes are pretty sparse.

Claude:

The morning is given over to preparation. The women of the village have been at work since before first light — great iron cauldrons of stew over open fires, flat barley-bread baking on hot stones, the last of the winter’s salted meat brought out and carved. Someone has found honey, and the smell of it warming in a clay pot draws every child in Stonetop to the Pavilion like moths.

Published:

The morning is given over to preparation. The women of the village have been at work since before first light — great iron cauldrons of stew over open fires, flat barley-bread baking on hot stones, the last of the winter’s salted meat brought out and carved. There is honey, brought from the Great Wood by a few clever hunters, and the smell of it warming in clay pots draws every child in Stonetop to the Pavilion like moths to a flickering flame.

Edited for worldbuilding specificity and the addition of a little color to the final sentence.

In its draft, Claude nailed the Padrig/Heledd moment.

Claude:

Four couples stand before the village. Padrig and Heledd are first — he in a clean tunic that Anwen suspects Seren had a hand in, she with wildflowers braided into her dark hair. Whatever Heledd thinks of the match that Cerys has made for her, she does not show it. She stands beside Padrig with her chin high and her hand steady in his, and when Garet asks if she takes this man freely, her voice carries across the green without a tremor. Padrig’s voice is quieter. Anwen has seen him face sorcerers and the walking dead with less fear in his eyes.

Published:

Four couples stand before the village. Padrig and Heledd are first — he in a clean tunic that Anwen suspects Seren had a hand in, she with wildflowers braided into her dark hair. Whatever Heledd thinks of the match that Cerys has made for her, she does not show it. She stands beside Padrig with her chin high and her hand steady in his, and when Garet asks if she takes this man freely, her voice carries across the green without a tremor. Padrig’s voice is quieter. Anwen has seen him face a storm-drake with less trepidation in his eyes.

One small change, again adding some specificity — in the story, Anwen has never seen Padrig face the walking dead. Claude was reaching for something fearful and grabbed two examples — a sorcerer, which Padrig has fought (though never directly faced), and the walking dead, which he has never had to tangle with. I cut both and replaced them with a storm-drake, which they have faced together back in Session 5 (and Pad again on his own more recently in Session 13).

This small change points to a broader element of writing craft that writers who want to use LLMs as part of their process need to navigate. All genre writing involves weaving elements of the genre canon and traditions with novel elements that make the work your own.

In the course of writing a chapter, you must make hundreds of microdecisions about when to evoke something that’s present across many genre works (the flowers in Heledd’s hair) and when to envison something that’s only present in your specific work (Anwen’s memories of seeing Padrig face down a storm-drake). Claude is decent at making connections to the broader genre — it’s very well-read, after all — and often falters when trying to make connections to what’s come before in the work itself. This is one of the many reasons why, when you write something with an LLM as your partner, you have to take it the final mile yourself.

Opus ipsum, the work itself

There you have it. From a certain point of view — a point of view that I see as dominant in the writerly community — this is a disqualifying amount of LLM-content. It means that PTFO:Stonetop isn’t really my work, or it isn’t really ‘art.’

But I think it still is very much my work, and I think it’s art—not high art, but certainly some sort of art-flavored pop culture—even though an LLM helps quite a bit with the work. I think this because I feel the feelings when I sit down at the keyboard to work, and people feel them when they look at the work itself.

That, if anything, is the point of all this yap. There’s a lot of writing out there. The only way to tell if any of it is worth a damn is by reading it yourself and feeling something. You can’t outsource your taste to Pangram — only the work itself, opus ipsum, has the answers.

Case in point — I fed the passage from 16.1 into Pangram:

Can’t trust anyone these days!

Certified 100% human. Different passages have different results — feel free to try it yourself! But regardless of the result, it can’t tell you if it’s worth a damn. Only you can do that.