On Tuesday, OpenAI CEO Sam Altman posted a short piece of fiction to X. The author of the story, about a character named Mila who turns to AI for comfort during an experience of grief, refers to artificial intelligence as “nothing if not a democracy of ghosts.”
Who — or what — better to write such a phrase than an AI ghost itself?
You see, the story was written by a new AI model that is “good at creative writing,” according to Altman. He prompted the model with the following instructions: “Please write a metafictional literary short story about AI and grief.”
we trained a new model that is good at creative writing (not sure yet how/when it will get released). this is the first time i have been really struck by something written by AI; it got the vibe of metafiction so right.
— Sam Altman (@sama) March 11, 2025
PROMPT:
Please write a metafictional literary short story…
Well, the model certainly delivered … something. To find out if the story hit the metafiction mark, we asked bona fide authors, writing instructors, and literature professors across the Bay Area to share their thoughts.
Dave Eggers, author of dozens of books, founder of literary magazine McSweeney’s, and Pulitzer Prize finalist for his memoir “A Heartbreaking Work of Staggering Genius,” was not having any part of it. “AI can cut and paste text stolen from the internet, but that’s not art. It’s pastiche garbage that would fool only the most gullible,” he wrote in an email to The Standard.
Eggers believes “there are so many good uses for AI,” but creative writing is not one of them. “There is no such thing as AI creative writing. What AI does is mimic syntax. It’s a cheap party trick,” he wrote. “Art comes from the human soul and all we’ve seen and lived, all the pain we’ve known, all the beauty we’ve witnessed, all the lessons we’ve learned.”
Not everyone was as instantly critical of Altman’s AI story. Fiction writer and University of San Francisco professor Susan Steinberg wondered what she would have thought of the piece if she hadn’t known beforehand that it was produced by AI. “If somebody I knew had written it, I would’ve helped them through it, but there’s a lot I would’ve revised out of it,” she said.
The story’s overabundance of metaphors, to name one literary crutch, felt like a mockery of a certain “writerly style,” Steinberg said. (The Standard asked ChatGPT to count the metaphors in the story; at first, it said 17, then at least 21 and, at most, 38.)
“AI creative writing feels like it’s about a product,” said Steinberg. This misses the point of great writing, which is to produce something beautiful, anguished, and conveyed in a way that feels natural and authentic to its creator. “I don’t want the art I look at to be efficient,” she said. “I want it to be messy and human.”
Jenny Bitner, a novelist and writing teacher at The Writers Grotto, which offers workshops in SoMa, offered a small amount of praise for the AI fiction, describing it as “clever.” However, “there’s a sort of plot problem,” she added. “Toward the end, it’s making a twist, like it was never supposed to tell you that it was an AI creating the story. But it tells us that [in] the very first line.” If she were revising the story, she’d cut the first paragraph.
San Francisco writer and poet Selen Ozturk got through a few paragraphs of the short story before she scrolled away. “It’s like slop,” she said, “and is basically as cliched as any 13-year-old’s diary.”
Ozturk, 24, doesn’t feel threatened by OpenAI’s innovation. If this is the best it can do, she feels writing professionals need not worry. In fact, she would be “kind of offended if AI used my work, and this was the best it did.”
Among her community of creative writers, the sentiment around AI is that of “memeified scorn” rather than an actual sense of threat. “The advent of photography didn’t demolish painting, and digital photography didn’t demolish analog,” Ozturk said.
Though generative creative writing hasn’t been a focus for AI companies, the inputs that build the large language models powering chatbots have garnered attention and controversy. Creative writers have sued OpenAI and other companies for copyright infringement because their AI models were trained on protected material. OpenAI has admitted as much, saying in 2023 that “it would be impossible to train today’s leading AI models without using copyrighted materials.”
Laird Harrison, a novelist and instructor at San Francisco State University, said that while AI models have generally improved, for the purposes of creative writing, their performance has declined. “They use this process called human-reinforced learning,” he said, in which engineers train models to steer clear of offensive language and violent or sexual imagery. In the process, AI models became “less interesting and less likely to come up with something surprising or original.”
Harrison, who taught a class on writing with AI at The Writers Grotto, said that prompting ChatGPT to write a creative piece usually generates a “very trite and kind of predictable” response. “What’s interesting about this particular iteration is that it’s better at creating interesting images,” he added, referring to OpenAI’s latest model. But character development — the stuff that comes from the writer’s human experience — is lacking. What does the character care about, fear, want? From the AI-written excerpt Altman shared, it’s impossible to tell.
“I don’t find myself caring at all about these characters,” Harrison said. “They don’t have any complexity to them.”
Even with the obvious downsides, Steinberg can’t help but wonder if one day publishers will favor pieces generated by AI because of the value and efficiency. As a human writer, she is dreading that day. “Just because we have a technology doesn’t mean we have to use it,” she said.