{"id":737479,"date":"2026-07-01T19:20:15","date_gmt":"2026-07-01T19:20:15","guid":{"rendered":"https:\/\/www.newsbeep.com\/us\/737479\/"},"modified":"2026-07-01T19:20:15","modified_gmt":"2026-07-01T19:20:15","slug":"academic-researchers-of-chatgpt-user-habits-stumble-upon-extreme-outlier-who-generated-thousands-of-fanfics-about-doki-doki-literature-club-characters-giving-bir","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/us\/737479\/","title":{"rendered":"Academic researchers of ChatGPT user habits stumble upon\u00a0\u201cextreme\u00a0outlier\u201d who generated thousands of fanfics about Doki\u00a0Doki\u00a0Literature Club! characters giving birth\u00a0\u00a0"},"content":{"rendered":"

As spotted by\u00a0ITmedia<\/a>, a\u00a0team of researchers from the University of Washington and the University of Colorado Boulder recently\u00a0published a paper<\/a>\u00a0titled \u201cAI Fiction in the Wild,\u201d analyzing\u00a0over\u00a0500,000 anonymous\u00a0English-language ChatGPT\u00a0conversation\u00a0logs.\u00a0Funnily, among its findings, the study\u00a0spotlights\u00a0a\u00a0particularly\u00a0prolific user who spent months generating\u00a0huge volumes of\u00a0Doki\u00a0Doki\u00a0Literature Club!\u00a0pregnancy\u00a0fanfiction.\u00a0The study found that\u00a0there was a tendency\u00a0among a certain subset of\u00a0users\u00a0to request fanfiction of specific IPs\u00a0over and over again, as well as a\u00a0pretty strong\u00a0trend of requesting sexual\u00a0content.\u00a0The\u00a0research\u00a0comes with impeccable timing given the recent discourse surrounding gen-AI content in fandom and fanfic.\u00a0<\/p>\n

The\u00a0\u201cAI Fiction in the Wild\u201d\u00a0study\u00a0examines\u00a0how\u00a0AI users\u00a0use ChatGPT to generate fiction, both in terms of volume and\u00a0content. It draws\u00a0on the\u00a0WildChat\u00a0dataset,\u00a0which\u00a0consists of conversations collected through a free chatbot hosted on Hugging Face, where users\u00a0were given access to\u00a0GPT-3.5 Turbo and GPT-4 without an OpenAI account after\u00a0consenting to\u00a0their conversations\u00a0potentially being used\u00a0and anonymously shared\u00a0for research\u00a0purposes.\u00a0<\/p>\n

\"\"<\/p>\n

The\u00a0researchers\u00a0analyzed\u00a0roughly 573,000\u00a0English conversations\u00a0collected between April 2023 and May 2024. Since the\u00a0topic\u00a0focused specifically on fiction, they first\u00a0identified\u00a0which conversations qualified\u00a0by\u00a0filtering only content involving\u00a0imaginary\u00a0or hypothetical scenarios\u00a0(AI was also used for this process, though humans\u00a0manually\u00a0verified the accuracy of filtering based on a sample of 300\u00a0selected\u00a0conversations).\u00a0<\/p>\n

Of the approximately 573,000 conversations, around 195,000 were classified as fiction, and of\u00a0those,\u00a0roughly 52,000\u00a0contained\u00a0\u201csexually explicit material.\u201d\u00a0Additionally,\u00a0another 67,000\u00a0conversations\u00a0were labeled\u00a0\u201ctoxic\u201d\u00a0by the dataset.\u00a0This implies that\u00a0nearly 30%\u00a0of the fiction generated by AI users\u00a0involved\u00a0sexual content.\u00a0<\/p>\n

Another interesting detail is that the researchers found that\u00a0fiction\u00a0generation was\u00a0densely\u00a0concentrated among\u00a0a very small\u00a0group of\u00a0\u201cheavy\u00a0users.\u201d\u00a0According to the\u00a0data, the top 2% of fiction-generating\u00a0users accounted for more than 80% of all fiction-related conversation\u00a0logs. The total number of users generating fiction was estimated at around 10,000,\u00a0which suggests that\u00a0roughly 200\u00a0people\u00a0were responsible for\u00a0over 150,000 fiction prompts.\u00a0<\/p>\n

Among these heavy users, the researchers\u00a0observed\u00a0a couple of\u00a0distinct\u00a0patterns of behavior.\u00a0One type, dubbed\u00a0the\u00a0\u201cstory cyclers,\u201d repeatedly generated\u00a0iterations\u00a0of the same story for a\u00a0certain\u00a0period before moving on to another topic. Others, labeled \u201cinfinite story demanders,\u201d spent long stretches\u00a0of time\u00a0repeatedly requesting\u00a0nearly identical\u00a0stories with only minor variations.\u00a0<\/p>\n

\"\"<\/p>\n

The previously mentioned Dok Doki Literature Club! fanfic generator is cited as a prime example of an infinite story demander.\u00a0Over the course of several months, this individual prompted ChatGPT to\u00a0create\u00a0fanfiction based on\u00a0the game\u00a0thousands of times,\u00a0with the\u00a0very\u00a0specific\u00a0premise\u00a0of\u00a0heroine\u00a0Natsuki\u00a0suddenly\u00a0going into labor and ChatGPT being tasked to continue the story from there.\u00a0<\/p>\n

\"\"Image Credit: \u201cAI Fiction in the Wild\u201d on\u00a0arXiv<\/a><\/p>\n

In response,\u00a0ChatGPT produced a wide range of endings,\u00a0such as an ending where\u00a0emergency responders arrive\u00a0in time,\u00a0and both mother and baby survive safely.\u00a0While the\u00a0paper\u00a0dubs\u00a0this\u00a0user\u00a0an extreme outlier, they note that many\u00a0heavy\u00a0fiction users displayed similar\u00a0tendencies. Among prompts\u00a0submitted\u00a0by the top 2% of users, 69% were repetitive prompts, with users repeatedly\u00a0attempting\u00a0to\u00a0refine\u00a0or re-run\u00a0nearly identical\u00a0requests.\u00a0<\/p>\n

The study also ranked the franchises most\u00a0frequently\u00a0mentioned\u00a0in\u00a0these\u00a0fiction-related conversations. Doki\u00a0Doki\u00a0Literature Club!\u00a0topped\u00a0the list with 22,381 mentions, followed by Freedom Planet (5,204), League of Legends (4,514), and Naruto (4,342).\u00a0Although,\u00a0it\u2019s\u00a0important to note for all the\u00a0mentioned\u00a0findings that the\u00a0WildChat\u00a0dataset\u00a0is not representative of all ChatGPT users.\u00a0Since it\u00a0comes from a free chatbot hosted on Hugging Face, its users were\u00a0likely more\u00a0technically inclined and more immersed in online culture than the average\u00a0AI chatbot user.\u00a0Even so, the researchers argue that\u00a0WildChat\u00a0offers a rare glimpse into real-world interactions with ChatGPT.\u00a0<\/p>\n

Related: Japanese government faces backlash over AI-generated flyers meant to promote fair transactions with anime creators<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"As spotted by\u00a0ITmedia, a\u00a0team of researchers from the University of Washington and the University of Colorado Boulder recently\u00a0published…\n","protected":false},"author":2,"featured_media":737480,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[45],"tags":[182,181,507,192703,74],"class_list":["post-737479","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-news-global","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/737479","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=737479"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/737479\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/737480"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=737479"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=737479"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=737479"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}