Sam Altman’s AI Short Story Sucks, and He Doesn’t Know Why

OpenAI CEO Sam Altman’s job is to talk up the promise of AI, whether its capacity to destroy the job market, destroy the world, or merely how well ChatGPT can put words in front of other words. Like most tech bros who read science fiction and imagine themselves creating the “torment nexus” of their dreams, Altman doesn’t comprehend what makes a good story. OpenAI is working on an AI model meant to write fiction as well as a human. Too bad the writing—while more verbose—still sucks.

“Like a server farm at midnight.” That line alone would get you laughed out of grad school. It certainly wouldn’t turn heads from anybody who regularly reads. But if a large language model writes it, can we say it’s impressive? Altman likes to think so. He posted this AI-written short story to his X page Tuesday, saying, “This is the first time I have been really struck by something written by AI.”

If a human wrote this, I could at least analyze its intent. With an AI, there’s none. I’m struck by how awful the writing is and how willing AI evangelists are to pass this stuff off as profound. Does Altman think writing a metatextual piece is somehow harder than writing straightforward fiction? If so, then to compensate, the AI becomes more florid and prone to excess and purple prose.

Let’s pretend we’re a professor of creative writing, and we have to grade this. In the second paragraph, we’re already too far into the weeds. I get it, you’re AI, but the “blinking cursor” motif is as overused as the “it was a dark and stormy night” line. Next, the line “Mila fits in the palm of your hand, and her grief is supposed to fit there too.” This is wordy, and it doesn’t follow from the rest of the paragraph. We haven’t established Mila at all, other than her name, but we are already supposed to assume she’s grieving.

Being metatextual doesn’t grant you more leeway to throw characters around like a kid playing with action figures. It gets worse. “I don’t have a kitchen, or a sense of smell. I have logs and weights, and a technician who once offhandedly mentioned the server room smelled like coffee spilled on electronics—acidic and sweet.” What the hell does that mean? For one, coffee doesn’t smell acidic. It may taste acidic; however, the only electronics that seem to be shorting out is this AI that is merely stitching together fragments of other people’s written work.

It goes on like that. Just because it uses big words and more of them doesn’t mean the text is meaningful—quite the opposite. It becomes muddled and vague. The text is offputting in a way that becomes clear the harder you analyze the text.

“Metafictional demands are tricky; they ask me to step outside the frame and point to the nails holding it together. So here: there is no Mila, no Kai, no marigolds. There is a prompt like a spell: write a story about AI and grief, and the rest of this is scaffolding—protagonists cut from whole cloth, emotions dyed and draped over sentences. You might feel cheated by that admission, or perhaps relieved. That tension is part of the design.”

I’m sorry, ChatGPT, but you can’t cop out. You can’t step outside “the frame” to soliloquy about the nature of metatextual writing on a whim. It’s trite. There are lines in the AI-written piece that echo something I could imagine a human would write, but pretending profoundness does not make a story cohesive. You don’t need big words to make text literary. Do you imagine Ursula K. Le Guin’s Earthsea Cycle is somehow less profound because it was written with young readers in mind?

OpenAI is working on multiple updates to its LLMs and reasoning models, but all signs point to them losing steam. Earlier this month, the company launched ChatGPT 4.5 exclusively for paid ChatGPT subscribers. The company claims this model has “emotional intelligence and creativity.” But how do you judge an AI’s creativity? If you ask ChatGPT to write a poem as TechRadar did, can you tell which version is GPT-4o and which is GPT-4.5?

OpenAI hopes GPT-5 will also integrate the company’s o3 reasoning model. This should make the AI better at checking its work (emphasis on “should”). That model should arrive sometime in the first half of this year. We doubt more reasoning capabilities will make a big dent in its “creative” output.

All this does is incentivize people who don’t know how to write to pass off AI-generated slop as their own. We saw the impact of AI back in 2023 when a hoard of grifters flooded the submissions page of Clarkesworld magazine with trash to make a quick buck. On Amazon, a slew of AI-generated books—some fully plagiarized from other works—clogged up the streams of people trying to promote their self-published work. There was so much Amazon has tried to get submissions to label whether AI created them.

When Altman promotes AI’s literary talents, he’s trying to create a new market for ChatGPT subscriptions by promising uncreative people they can take the reigns from the literary “elite.” But the thing is, even if you imagine a human created this, it’s still trash. Knowing AI created it, it’s doubly trash. Nothing in there, no speck of creative intent, makes AI drivel worth reading.