Setting an example for authors to disclose their AI use

In light of the new rules, and as a long-time and reasonably known author, I want to set an example to encourage authors to disclose their AI use.

In spring of 2023, when ChatGPT started to become popular, I experimented with it and created two series of stories heavily assisted by AI. However, I am a writer above all and even though it was new to me, I only ever saw it as a tool and did not delegate any creative decisions to it. The creation of those stories involved almost a thousand prompts altogether, on ChatGPT 4 for non-sex scenes and a jailbroken clunky ChatGPT 3 for sex scenes. There were prompts to generate text yes, but most of the prompts involved brainstorming with it to create the characters, the setting, basically a very large writing bible before even starting on the first line. Another large part of the prompts were as a thesaurus, and as questions like “Do we say came into or came inside usually?”, “What would be a nice location in Maine for this kind of house?”. All of the text was heavily heavily edited because what came out of those generations was often awful. All the twists, the humor, the plot, the themes, etc… came from me, I would never trust an AI with that. I wrote one chapter while I was in the hospital after a heart attack.

Both series are about 30000 words each (divided into 4 chapters each) and I spent about 120 hours on them total, for 60000 words. Overall, it took me about twice the effort I would have spent writing them manually from scratch, so low effort my ass. Still, they gathered dozens of favorites and comments and the first chapter of Human Trials earned a silver Mind Control badge. They stand as an example of what a writer who treats AI only as a tool, doesn’t delegate it any aspect of creativity, and certainly doesn’t delegate the effort of doing the real final writing to it, can achieve.

I have the pages and pages of chat logs still in archive, and I might post in the forum to give more details on how it was created for forensic interests.

They’re still not my best stories, they were just an experiment, and they are full of AI Patterns that make me cringe today. I had taken them down a few months before the AI rules came in effect, I will now reactivate them with a proud “Contains AI Patterns” tag AND a disclaimer in an effort to encourage other authors to disclose.

3 Likes

Oh, just reacting to something said in the other thread, only 5% of those 120 hours was spent crafting or refining prompts for the AI. 95% of the work and effort was writing and editing directly the words in the story. So the work was certainly not finding the best way to describe a fantasy so the machine would spontaneously poof it into existence.

I don’t think many, if any, authors using AI do it anywhere close to the way I did with those 2 experiments. It’s too much work and it requires strong writing skills from the get go. But those experiments establish a spectrum of AI use from the most lazy “write me a scene with a guy hypnotizing a fireman into stripping for him” to the very involved process I used in my experiments where the machine was used mostly for inspiration or sophisticated queries, without any creative delegation at all.

2 Likes

Great idea, maybe we can (internaly) use the story [retracted] as an example and you comment for us, why we handle this as a ban?

Pup, this discussion is public, not internal :slight_smile:

People can’t read the unpublished story and we don’t want to publicly shame an author either.

I’ll edit your post to remove the link. I think you missed the fact that this discussion is public.

There’s actually nothing different in the detection for bans vs tagging-encouraged. It’s all the same signs and patterns as outlined in Soren’s document, The difference is in how much it permeates the entire story, almost like every paragraph shows signs of it. If there’s hesitation, I have access to high quality paid detectors that are much more reliable than the free crap (still never perfect), and these stories ping consistently in the 90%-100% likely AI ranges in 2+ detectors. They’re rarely needed. But basically, all the patterns highlighted by Soren, but on steroids.

Also, these texts show clear signs of not having a human touch at all guiding the writing.

I mention more here: The rules for AI involvement in stories - #143 by Mafisto

2 Likes