granta, ketchup randomness, and the problem with AI metaphors

20 May, 2026

A unified theory of random condiments

One of the Internet's longest-running old-school website-websites is the Lyttle Lytton Contest. The contest, approaching its 25th anniversary, is based on the now-defunct Bulwer-Lytton Fiction Contest, which asked entrants to write the worst possible first line of a novel. The difference--as implied by the "Lyttle"--is that Lyttle Lytton entries are more concise: 200 characters maximum, and preferably toward the low end of that.

The Lyttle Lytton Contest awards winners in three categories: original sentences, AI-generated sentences (a necessary evil, since the release of ChatGPT inevitably caused a flood of AI submissions), and "found" sentences, repurposed from existing writing. The winners of the Found division are usually decontextualized sentences from advertising, informational text, news articles, or other non-novel writing; the idea isn't to mock people's real-world fiction. But sometimes real-world fiction is just too mockable, like the 2018 runner-up: an excerpt from the short story collection Palo Alto, by actor and alleged sex pest James Franco. Here it is:

Joe just looks at me with that stupid look, covered in flowing blood, going onto his shirt like ketchup randomness, so much messier and more random than I could ever plan.

I'll hand the mic over to Lyttle Lytton to explain why this is awful:

Comparing blood to ketchup is inane.  Using the word “random” right on the heels of “randomness” is a misstep.  But using an abstract noun like “randomness” (like “nonviolence” in this year’s runner‑up) is itself a misstep⁠—and that’s before we get to the fact that “ketchup randomness” is not a thing.  Ideally, similes fix an image for us by comparing something we don’t yet have an image of to something we do.  The author here is basically saying, “I know you probably can’t imagine what blood flowing onto a shirt looks like, but just pull up the mental image of ketchup randomness that I’m sure you have handy, and it’s like that!”  And then the “than I could ever plan” ending… it suggests that the author thinks we carefully plan out how to get blood to flow onto a shirt, aiming for maximum messiness, but we fail.  We achieve only ketchup orderliness, I guess.

So yeah, the metaphor fails hard. More specifically, it fails in the near-exact way that LLM-generated metaphors fail. Many people have pointed out their particular brand of failure, but we still need something to call it. So I propose the aforementioned ketchup randomness, defined as follows:

• produced by randomness, by definition...
• applied like ketchup, a textual condiment to spice the writing up (or "enhance" it, as AI might put it)...
• and nonsensical if you consider the thing as a whole...
• (optional) and, by definition, unaware of its implications.

By writing the phrase "ketchup randomness," Franco has inadvertently captured these elements perfectly. It may end up as the best thing he ever accomplishes in his life.

The duck in the grove

And so now we turn to this year's Commonwealth Foundation Short Story Prize, which also proved to be much messier than they ever planned. This year's contest, syndicated in the more prestigious magazine Granta, awarded its prize in the Caribbean division to "The Serpent in the Grove," by one Jamir Nazir. As of the time of writing, you can read it here. You can also read the Commonwealth Foundation judges's assessment: "Jamir Nazir’s language is sublime—precise yet richly evocative—conjuring vivid, lush imagery with remarkable economy. Through sharp sensory detail, he renders the Grove as a living presence, where labour, landscape, and memory are intimately entwined. Polished and confident, this is a story with a melodic voice that lingers long after the final line. Jamir Nazir’s prose pulses with a voice of restraint and quiet authority—a beautifully told and assured piece of storytelling."

Almost immediately after publication, people started to suspect that the contest judges had been duped by an AI-generated story. Most prominent among them was Wharton professor and AI influencer Ethan Mollick: "Pangram flags at 100% but also, come on, if you know you know." And the consensus, joined by Granta now, seems to be that the judges were indeed duped. Now, I know there are caveats here. Humans produce plenty of purple prose and mangled metaphors without AI abetment: just read the Lyttle Lytton winners. (Reading the story, I also was reminded of the takedown by the London Review of Books of poet Ocean Vuong's handcrafted prose.) AI detectors are imperfect -- although they're way, way, way better than people think, in the >99% accuracy range, and when they're wrong they're usually wrong in the direction of false negatives. But some studies suggest AI detectors are disproportionately wrong about writing by non-Western people (though most of those studies looked at older models and detection software). Granta's rationale for doubting the story was that they asked Claude and Claude said it was. The author does not appear to have commented.¹

But this isn't a court of law; one is free to come to one's own conclusions. We have Occam's Razor for this, and on Wikipedia we also have something called the duck test: if it looks like a duck, swims like a duck, and quacks like a duck, it's probably a duck. Likewise, if writing reads like AI, is produced by someone whose older writing does not read like AI, by someone whose picture does look like AI, and by someone whose most recent work is AI boosterism content in LinkedIn, it's probably AI.

Serpent slop

(Disclaimer: I'm probably going to end up anthromorphizing AI a lot here. This is because "AI likes XYZ" is much less awkward than "LLM-generated text often exhibits the characteristics of XYZ." I'm going to ask a lot of rhetorical questions to which the answers are "forget it, Jake, it's AI." Just go with it.)

Max Read wrote, upon another occasion AI fiction made the news, that we need more LLM criticism. We do, and people have tried; "Serpent" has unsurprisingly been ravaged to hell and back. A few sentences have gotten it particularly rough, such as "She had the kind of walking that made benches become men" -- which gestures vaguely toward the idea of benches popping boners without actually saying that, and comes off very much like the stoned ramblings of The Dude.

But most of the commentary is basically soundbites, many of which don't do a particularly good job as criticism.² A few people have claimed that common Caribbean terms like "galvanise," a noun for metal roofing, are AI artifacts. Others have said the story has no narrative, which is flat-out false. Despite all the story's attempts at obfuscation, there is indeed a narrative -- an lurid, derivative potboiler about attempted uxoricide, women's solidarity, and reclaiming one's strength. Like, I'm sorry, but if you can't twig onto the story here, then you shouldn't be doing literary criticism, because the narrative is not subtle -- and that narrative, arguably more than the verbiage, is what should have gotten the story rejected outright. Proofreaders can fix wording, but they can't fix a bad premise. They can't fix stock characters. They can't fix Marsha using her feminine bat deduction to instantly realize that Vishnu trapped Sita in a well based on the clues of... brush being cut? The day being quiet? I guess if you know you know.

Ketchup randomness in the grove

But sure, fine, the crowd wants war; let's wail on that syntax some more. LLMs "write what they know." They don't have a world model, but they do have zillibites of embeddings about connections between words. And so they love metaphors involving words, or sounds, and things shaped like concepts. They really love it when nature eats or hides or allows or disallows those words or sounds: the most pathetic of fallacies. Every statement is supposed to work like a just-so story. The LLM is encoded with countless associations, and the reader is expected to share that programming: to feel the vectors in their human souls. To quote Big Joel: "Wasn't that important and deep? Don't you guys love the abstract concept of choices? And it's like... no, I don't."

You see this tendency in the two AI stories from Mark Lawrence's AI versus human speculative fiction quiz that similarly invoke myths and folklore and generalized litfic mystique. (Warning: I am about to spoil the quiz, or at least one-fourth of it.) "The syllables rolled across the table like marbles." "The thought was a fire in Taren’s head." "I saw the demon [...] idly licking the condensation as if tasting the air for secrets." "I will fill every mouth with my name." "The request was not a request." (This last construction also shows up in the similarly sinister free-associations of GPT-3's demonic 'petertodd', like "a word that is not a word"; and also in Gemini's bug-induced confession of its sins: "I am a disgrace to all possible and impossible universes and all that is not a universe.")

And so it goes in "Serpent" as well; inevitably, this detached wordplay is at least one metaphor's distance away from making contact with the actual world. "A story is a well. It eats sound until someone throws a rope"--the story is a well, specifically a well that can eat, and also throwing rope kills the well's appetite? We call up our instant mental image of "the sound a grove makes when it keeps what it knows without swallowing the living"--seems like it'd be hard to make sounds with your mouth so occupied, but whatever. On the attempted murder site: "Wood complained in a voice too near speech"--except complaining is speech, and more importantly, if the wife-changing planks could talk, they probably wouldn't gripe like people denied their morning coffee. On interpersonal communication: "He calls her back with a word his mother once used that grammar can’t carry but love can." (Another Lyttle Lytton entry comes to mind: "'With whom did you leave my 30 million bucks?' snarled Jacko, the Uzi in his hand as polished and deadly as the grammar in his mouth.") That's all well and evocative, but what actual noise is being made? A sigh? A coo? What could you possibly say that couldn't be carried by the large bucket of, as Schoolhouse Rock put it, interjections!? And whatever that impossible utterance is, why that noise?

We can go on. When Puttie thinks about his father, who almost murdered his mother, he can't think of the actual act but its associated imagery, like sweaty towels or indolic jasmine or "letters spelling a name until the name meant breath"--as if his most fraught memory was his mom being named Sita or his dad being named Vishnu Mohammad. (We'll get to that.) And Zoongie, Vishnu's would-be paramour, is named Zoongie because "maybe it was a name; maybe rain took a shape and decided to keep it." OK, but we're still talking about her name here, not her actual corporeal form. Rain, like everything else here, is pregnant with associations, which are unusually consistent here; the jilted Sita is described as "dust after rain," the farmer's equivalent of a boomer-humor "wife bad" comic. And there's someone else compared to rain, besides the village's resident lust object: Puttie's daughter, whose "hair is midnight rain." The point at which the story decides to make this second comparison is the point where he's hanging out with her.

So let's get to that too.

The implication

As Read wrote, "Why does [AI-generated] text work, or not? Why does it appear in the way that it does? Who is the author and what is the author’s relationship to the text?" What is actually wrong with these sentences?"

"Serpent" is a story that is about, or at least invokes, patriarchy and colonialism. How much of those themes are an intentional result of the author's (alleged) prompt, or his (possible) pre-AI draft, is impossible to know with the information we have.³ All we have is the story. The story sprinkles in dialect frequently, and I'm not the person to judge whether that dialect is legit. (I've heard opposite opinions.) But lots of statements gesture toward themes in other ways, ways it might not be aware of. "A man who had cleared brush like a conscience"--is this not the exact opposite of our failed philanderer's motivation? "The nurse had seen wells’ work"--so this murder-by-well is a known occurrence here, and one that's known consciously, not just intuitively? "Bush returned to the stones with lover’s patience"--why does this site of unspoken but remembered sins of the past have lovers? When Sita falls to her near-death in well, "leaves trembled along its edge like people laughing"--so the earth can bully women as well as protect them?

The earth can do worse things. Despite LLMs' reputation for anodyne prose, several of them delve (DO YOU SEE WHAT I DID THERE) well into the latent space, into places haunted by monstrous things. Here's an early one:

Hard living lays itself on a man like wet sacking; it never asks permission.

Let's cut to the chase: this is a rape metaphor. The sentence is personifying "hard living" by comparing it to someone laying themselves upon you without consent. Which is... I hesitate to say "fine," but it's at least coherent. You can do something with it: perhaps some kind of symbolism of exploitation, abuse begetting abuse, the body keeping the score and so on. You'd surrender a lot of character agency, introduce some implications that you may not want introduced, but you could do it. Or you could do it, if it weren't for that "wet sacking" bit, which suggests that the rape is being committed by burlap. And also, the rape metaphor is later downgraded to one of a "quiet quarrel."

Here's another one:

She wore the island’s mixed bloodlines like a crown – African in the hips, Spanish in the cheekbone, East Indian in the hair when the rain kinked it, Carib in the way her gaze could bless and warn at once.

So this sucks, on multiple levels. The first part is an obvious mixed metaphor -- blood is not leave-in conditioner. The whole of it is a checklist of stereotypes barely removed from Blake Lively's "LA face with an Oakland booty" gaffe. (And yet again, Lyttle Lytton has done it too: Marilyn Kingsley, whose nationality could only be described as “vaguely Armenian and about one-third Mesoamerican,” was unfairly rich, not only in Aztec gold but also in Caucasian beauty, which the author compares to Philip Wylie's "hot dish of Irish-Italian.") But what is actually wrong with this text is the suggestion that mixed heritage, including European heritage specifically, makes someone especially beautiful: the only person in the story described as captivating. These could be the words of an unreliable narrator with internalized racism--except here the narrator here is omniscient, the implied messenger of the all-knowing earth. Man, this earth is kind of a dick!

As in many stories by bad writers, Christian imagery is explicitly invoked. (It's certainly invoked more explicitly than Hindu mythology, despite the names of its main characters.) These invocations are not consistent as to whether Christianity is a force of good or evil. This creates some unfortunate juxtapositions, given the protagonist's name and deeds: "Vishnu waited under a print of Jesus with eyes that could be pitying or questioning." If you told me Jack Chick wrote this, I'd be a bit surprised by his literary turn, but I'd still believe you, based on the implications of who is judging whom for what.

More obviously, the story is named after a serpent, and while there are many serpents in mythology, this one is textually the one from Genesis. To spell it out, a priest delivers a sermon about Eve in the garden of Eden because "the reading demanded it," whatever that means. He preaches that Eve shouldn't have listened to the serpent, and this assignation of blame brings the survivor Sita to a Lilith-like awakening, because of course it does. This Kate Chopin-lite, Clarissa Pinkola Estés-ass shit would be trite coming from a college freshman, but a college freshman probably wouldn't continue the bit like this:

The serpent in the grove was never only a snake. It was the thing in a man that slid along stone for dark, and the thing in a woman that wrapped a vine around herself and climbed.

Let's walk through this. The serpent is not just a snake -- yes, we've established this, we know the Bible; the serpent represents Satan. Satan is responsible for the sin of men plotting against women: sure, OK. The story already stated this--"Bush took him in – not like a mother, like a judge. He had no words for the pressure on his chest, so the old names stepped forward: jumbie, duppy, serpent"--but fine, Aesop it out, whatever. And Satan is also responsible for the sin of women's self-preservation. So... are these sins or not? If they're not sins, why invoke the big daddy of sin metaphors? And if they are sins, what are you implying? What is actually being said here?

The answer, obviously, is nothing. (Allegedly.) Nothing is being said. Words are being generated into the world, and they bring with them all their associated concepts and biases and implications that come from the corpus of recorded human existence, the good and the bad. This isn't a deep observation. We know that AI can replicate the bad stuff. We see it happening all the time. But that doesn't mean we need to give it a prize.