GoKawiilThe big news in the world of writing today is the controversy over the award of a Commonwealth Foundation Short Story Prize to a story called “The Serpent in the Grove.” The piece was almost certainly co-authored by AI.
As of this morning, the magazine that published the piece (the prestigious literary journal Granta) has not issued a retraction. Rather stunningly, in fact, Granta has just issued a statement about the affair that cites Claude as an arbiter of whether the story was AI-written or not!
More on the question of trust and experience later. Suffice to say that it does not take an AI-detection tool to spot the obvious ChatGPT-isms in the story.
The dead giveaway is the repetition of bizarre figures of speech. Mixed metaphors which sound nice at first glance, but slip away from meaning like an echo chasing itself off a cliff. Similes that catch in your mind like river trouts tangled in the roots of a redwood tree. Literary flourishes that thicken the air’s tang with their… ok you get the idea.
AI systems are especially given to talking about hums and other ambient sounds like static, as well as ambient environments (water, air, ozone). These are frequently pushed up against “earthy” words (tang, belly) and ennui-laden emotional states (longing, forgetting, sadness). Once you notice the patterns, they’re impossible to miss.
Some examples from the Granta story:
…air clung thick as porridge skin: damp earth, woodsmoke, and the sour tang of fermenting cocoa… …his laughter like water over pebbles… …the air sweet with cane and forgetting… …People passing said they sometimes heard the noon hum if the wind was in a mood. Not every day. The day had to choose… …the hum loud as if noon had tuned itself…
This controversy is not yet finished, and will likely be repeating itself again, and again, in the months and years to come. The issue is not just that authors are submitting AI-written prose, but that judges are using language models to assess that prose. Anyone who has tried passing AI-produced writing to another AI tool (even in the context of coding — for instance, asking Gemini to read a plan for a new feature produced by Claude) can attest that these tools simply adore their own outputs.
For instance, here is Gemini 3.1 Pro, the current top model from Google, reasoning about whether it likes the Granta story. What I find striking about this is that the features it identifies as the best aspects of the story are precisely the things that make me — as a human reader — think it’s astonishingly bad.
For instance, Gemini thinks the setting is “richly evoked” with well-drawn characters, whereas to me it feels like the story is floating in some kind of literary nether-region without any sense of place, character, or scene. And it finds the meaningless metaphors, like those highlighted above, to be “stunning.”
... continue reading