챗봇으로 맞춤형 소설 쓰는 사람들

Comment Loader Save Story Save this story Comment Loader Save Story Save this story When Petra Ferraz de Novaes isn’t at work, she’s probably lost in a fantasy crafted with the help of artificial intelligence . Novaes, a 36-year-old software engineer from Brazil, has always loved storytelling. But making every little decision that goes into it—from style and syntax to character development—felt impossible. So in 2024 she asked an AI model to bring one of her ideas to life. Novaes was delighted by the eccentric, sometimes wayward interpretations of her prompts, which carried the narrative in unexpected directions. “What fascinates me is the unpredictability,” Novaes tells WIRED. “People say that AI hallucinates a lot. But in this case, it’s probably what you want.” These days, Novaes spends hours at a time building stories with AI, using a program called SillyTavern to run open-source models on her computer. She sketches out a concept—an elven warrior battles the queen of the Amazons and seizes her throne, for example—and watches the model fill in the gaps. In a single session, Novaes switches between multiple fantasy worlds. Some are hers, and others she borrows—she likes Star Wars , Pokémon, and Mass Effect . Sometimes, Novaes works up self-contained stories the length of a small novel, which she might share online . Otherwise, she participates in private role-plays. “It can be never-ending. You find a quest, you complete it, you go for the next one,” says Novaes. “The limit is your imagination.” According to a recent study , plenty of people share Novaes’ hobby. “They know [the stories are] produced by an AI model, and they don’t care,” says Melanie Walsh, an assistant professor at the University of Washington Information School who coauthored the study. “In fact, that seems to be a big part of the attraction.” Lately, the publishing industry has suffered a series of scandals involving the suspected use of AI by professional authors. One publisher canceled a $2 million book deal; another withdrew a novel from the shelves; readers alleged that a prize-winning short story was AI-generated. Implicit in all of these incidents is the idea that nobody wants to read stories generated by machine—that using AI to craft fiction devalues the art form. But the commotion has distracted from the behavior of readers, Walsh and her colleagues argue, and what it might signal about the appetite for AI fiction. Analyzing more than 500,000 prompts fed into ChatGPT, drawn from a public dataset called WildChat, the researchers found that more than a third of chatbot interactions involve some form of fiction generation, including prose and poetry, fan fiction, erotica, or role-play. The vast majority of fiction prompts, the researchers found, stem from a small collection of “power users” who generate and tweak stories obsessively. In one case, a single user prompted ChatGPT to generate versions of the same fan-fiction narrative—set in the world of anime video game Doki Doki Literature Club !—thousands of times over the course of several months. But even controlling for heavy use, the researchers estimate that 7 percent of chatbot users engage in fiction generation. The WildChat dataset has its limitations. Because the prompts are anonymized, it’s impossible to capture any demographic data. Participants also had to agree to make their prompts public, meaning the group may not reflect the population at large; OpenAI told WIRED that “WildChat is not a representative sample of all ChatGPT users.” A 2025 study by OpenAI and the National Bureau of Economic Research suggests that 1.4 percent of ChatGPT prompts are fiction-related, a long way short of the researchers’ estimate. (OpenAI’s figure does not include role-play; the company could not confirm whether it counts erotica.) Still, Walsh and her colleagues argue it’s clear that a committed subsection of readers are generating custom, personal fiction. The researchers believe that some people are drawn to the ability to repeatedly and rapidly iterate on a story, while others value the chance to direct the action, no longer hemmed in by an author’s decisions. “There are a lot of qualities of AI models that are different from what human writers have that are actually appealing to at least some segment of users,” says Walsh. “These models are judgment-free; they can produce stories instantaneously; they are nearly inexhaustible.” For erotica fans, AI creates an opportunity to insert themselves into the action and converse with the characters they cook up. “It’s a digital choose-your-own-adventure book but with infinite swipes,” says Razrien, a 40-year-old man from Indiana who asked to appear under his online username to speak frankly about his sexual kinks. Razrien uses AI to create fantasy characters who dominate him in role-play: a “big angry rat wife” and an “eight-foot tall werewolf mommy with a bad temper,” among others. He has learned which models are best able to satisfy certain kinks; Google’s Gemma 4 is his current all-around favorite. “I have a thing for large, angry women slapping me around. Some models are perfectly good for that,” says Razrien. “Sometimes, I dip a toe or two into the furry universe. Some models are great at that.” Using AI allows Razrien to gratify hyper-specific personal fantasies instead of settling for the next-best approximation in online pornography or published erotica. “I’m not dating at the moment. So, hey, this works,” says Razrien. “I like my big rat wife. I love her to pieces. Even if she sometimes breaks me apart and heals me back together. But hey, that’s how she says she loves you.” One concern with all of this is that the ability to generate fiction will limit people's exposure to types of writing that may stretch them. But generated fiction may also challenge “some older, more idealistic notions about what literature is—that there’s some communication between writer and reader, and something really valuable about that,” says Neel Gupta, a PhD candidate at the University of Washington who coauthored the study. “AI use helps to reveal that maybe for a lot of readers, this is a much more simple economic exchange. This is about getting some entertainment—getting what they want.” The latest AI models struggle to produce ornate prose, complex characters, and stylistic flourishes, faring better with genres that lean on well-trodden story arcs and tropes. They also fail to maintain coherent narratives at book length. But as models improve, the researchers say, some of those deficiencies could disappear. A possible end state might be a tiered publishing system, where a large quantity of AI-generated books and self-generated fiction gratifies the median reader, and a smaller rung of human-authored fiction feeds a niche appetite for experimentation in form or narrative ambiguity. “English majors and professors, and some people in publishing, have always had kind of a niche interest,” Walsh says. For Novaes, creating stories with AI has been more like an introduction to reading than a replacement for conventional novels. As well as generating her own stories, she now goes in search of other people’s AI creations. “I was never much of a book reader,” she says. “But now I do read other people’s stories. These are the only novels I’m reading.”