I'm currently posting here a 10-chapter Novella called "Project Mnemosyne," written with the assistance of Chat GPT, but with extensive, long, detailed prompts and post-edits. It seems to be getting good reviews, even from people I consider some of the best posting on BC or anywhere.
I've done another story, very different in tone from the Novella but also rather different from anything I have written previously. Still, this has more input from me to begin with and more editing by me, too.
It's a short story called "Another Round". I'm probably going to post it later today. I'm trying to make sure everyone knows what it is because I don't want anyone to think I am forcing my experimental assisted fiction on them. Just read the tags and you will be forewarned.
Thank you,
Suzan
Comments
Tools
I like to think that there's a world of difference between using a tool and being idiotic with it. You use the tool, and it helps you do your job, that's fine.
-- Daphne Xu
My technique
My method is to spend time describing what I want to do with the story to the AI, like setting, characters, certain incidents and plot points. Then I write an extensive prompt for a scene. I examine the scene produced and perhaps suggest changes or corrections, get those then perhaps more discussion before writing an extensive prompt (200-300 words usually) for another scene.
Once I have all the scenes for a story or a chapter, I stitch them together and do at least two editing passes. If there are chapters, I stitch the chapters together and do at least another editing pass on the whole thing. When I edit, I'm correcting minor things, deleting some passages that don't work and writing new passages. Edited sequences typically end up 10-50% longer.
Even with all that work, using the AI allows me to produce 2000 finished words a day, compared to my normal 500-1000 (on good days).
Suzan
AI...
That is "fine", I feel like an AI...
That barely computes...
Gwen, the AI
Heh!
I know whereof you speak.
Suzan
I have used AI for one
I have used AI for one concept (multiple stories, same universe) and have found that ChatGTP does a better job than AuthorAI. That being said…I am against using the tool to actually write out a story and just running with the output it gives.
(I am not sure if that’s what is being done on Amazon) I feel the tool cannot write out a story in my voice (without heavily editing) but kudos to you if you are able to make the balance work.
Tools
I have two finished AI assisted projects. The first completed was Project Mnemosyne which is being posted, all ten chapters, one every other day, here on BCTS.
Mnemosyne was an idea I had had for years but had not tried to write because it seemed so dark and unlike the stuff I normally write which can be light and even silly. The AI tool allowed me to write prompts which put the problem of darkness on the AI instead of me. So I could concentrate on making sure the story flowed and made sense and communicated the theme and ideas behind the story. Also, Project Mnemosyne was at least a novelette and kind of exhausting to even think about.
The second complete story I have done with AI is Another Round, a slice-of-life sort of thing set in a bar, featuring a meeting between two women. I don't go to bars, and I warned the AI of that as we discussed the story, who the women were, what they might discuss, what the action might be and the result. Then I wrote a 200-word prompt, and the AI produced a 500-word scene. We did that six times and ended up with a 3000-word short story, which I edited strenuously, both as scenes piecemeal and as a complete story. I'll probably post it tomorrow or later in the week.
Suzan
I used AI once. I had to edit
I used AI once. I had to edit it heavily and still wasn't happy, The last chapter of seven years as a wife was AI and edited. It doesn't write like YOU do. I think if someone knows how you write, they can tell. But doing what you are doing is way better than purely AI stories.
Leeanna
Editing is certainly necessary
I wouldn't expect otherwise.
Suzan
Fiction is an appropriate use for AI
Legal briefs, judicial opinions, government funding decisions, health policy . . . Nope.
It’s not a question of what it can do, but what it should do. MIT Professor Joseph Weizenbaum’s Computer Power and Human Reason made this point well, 50 years ago.
Exactly
The Nope list is of things that require the kind of precision LLMs cannot provide.
Suzan