Educational researchers investigating how folks use ChatGPT to generate fiction unintentionally stumbled upon an excellent consumer repeatedly requesting tales that includes pregnant characters from Doki Doki Literature Club (as noticed by Automaton and reported on by ITMedia).
In Walsh et al’s “AI Fiction in the Wild” research, College of Washington and College of Colorado Boulder researchers analyzed the WildChat dataset. It needs to be famous that WildChat is a set of over half one million nameless English language conversations between customers and ChatGPT, compiled with consent to be used in analysis.
“Probably the most prolific consumer within the WildChat fiction dataset and the clearest instance of an infinite story demander started a fanfiction narrative within the Doki Doki Literature Club! Universe,” noticed the researchers on page 10 of their paper. This tremendous consumer prompted tales that includes numerous highschool ladies from Doki Doki Literature Club being pregnant. Natsuki unexpectedly going into labor was a typical theme throughout these prompts. Most often, the consumer gave the AI a protracted dialogue with a pregnant Natsuki that cuts off mid sentence, prompting the chatbot to proceed the story.
Though Natsuki’s being pregnant will not be a part of the 2017 visible novel recreation, fanfics and artwork that includes a pregnant Natsuki have been going round for some time. Plainly this consumer found by the research has a very intense obsession with any such fanfic.
“Whereas this consumer is a very prolific outlier, and we will’t say with certainty why they’re producing these tales, many prolific customers ask for a similar sorts of fiction in an analogous vein,” the researchers commented, noting that the common fiction-generating consumer entered repetitive prompts 42% of the time, rising to 85% among the many prime 10 most prolific customers, together with the Doki Doki Literature Club fan.
General, the researchers found that over one third (34% or 195,271) of WildChat’s 573,453 conversations had been used for fiction era, be it authentic novels, scripts, fanfics, or roleplay. The highest fiction varieties had been fan fiction (49%) and erotica (29%), taking part in into ChatGPT’s energy of with the ability to generate instantaneous responses to customers’ particular requests and area of interest preferences. Amongst these fiction conversations, 10% (57,724) had been specific and 17% (94,870) had been poisonous, suggesting a excessive share of tales that includes sexual or taboo content material, as flagged by OpenAI Moderation or Detoxify.
Walsh et al discovered that this AI generated fiction was dominated by a comparatively small group of heavy customers, just like the Doki Doki Literature Club prompter. Plainly the highest 2% of fiction-generating customers prompted 80% of those conversations with ChatGPT, suggesting that round 200 folks had been accountable for a whopping 150,000 fiction prompts.
Amongst these tremendous customers, the researchers recognized “infinite story demanders,” specifically those that repeatedly request the identical extremely particular tales with solely slight variations, typically for months on finish. Additionally they recognized “story cyclers,” who generate variations of the identical story again and again for some time earlier than shifting on to a unique matter.
The most typical sources general for AI-generated fanfics within the WildChat dataset had been video games and manga, together with Naruto, League of Legends, Freedom Planet, and, in fact, Doki Doki Literature Club.
Verity Townsend is a Japan-based freelance author who beforehand served as editor, contributor and translator for the sport information website Automaton West. She has additionally written about Japanese tradition and films for numerous publications.