AI writing tools have revolutionized content creation, but they come with a unique challenge: the tendency to produce formulaic phrases and patterns that experienced users can spot immediately.
I've been trying my hand at this using ChatGPT, because you can use the latest model for free, and can keep the session open, even as you run into usage limits. It's been fun and interesting, for a story in my mind for decades. I need to prompt a lot, and sometimes direct it to specific parts of the source material I uploaded, but I find it satisfying so far. Grok 3 is free but left me unsatisfied. You've written very positively on Claude Opus, but it costs, and I'm on a budget.
So far, I don't think my story has fallen into this uncanny valley, perhaps because of its unusual subject. It even suggested some unexpected phrases I found apt.
What other models do you recommend that might be good for creative writing?
I have tried using Grok for some stories and one thing thats disables it is that Grok for each output always try to distill as much information as possible from previous pages into the next page. So Grok will always be telling the reader that they are in the "Blood Temple room" although the reader already knows this from 5 pages back.
Will promoting ai to write in a certain author’s style, or a mix of author’s styles, alleviate the problem? Such as if I prompted the ai to write a fantasy story in the style of Tolkien and Lewis.
I tried this out and verified what you are saying. Would it be different if you built a local ai that only understood the two example authors? In other words, if I scanned all the works from two authors into the local ai would the outcome be different?
I’ve noticed that AI when creating dialogue between two characters is often very abrupt, like it’s running down its checklist of bullet points that you provided in an outline. This makes it feel very unnatural as people tend to ease into uncomfortable topics.
Also, most surprising to me is how AI gets things wrong like a real person does. Sometimes events happen out of sequence even though I provide a timeline/outline of the sequence. And sometimes it “forgets” basic facts about the character/events.
The idea that authors will become stupid using AI — I’m not convinced. It simply makes things faster, but the author still needs to screen, edit, and be prepared to rewrite absolutely everything.
Biggest one is AI getting confused after a few rounds of editing. It can start to emphasize or add things completely irrelevant to the story's design.
Usually happens when discussing the work and potential futures. "No, AI, I do not want you to continue utilizing those hypotheticals tgat aren't going to be included in tje final draft." The solution is to start a new session to get it back on track.
I have yet to produce a "fully AI work" as the current process is to write the story and use AI for revising purposes while I make the adjustments by hand.
It does limit AI-isms as the AI is "airgapped" from the text.
I've been trying my hand at this using ChatGPT, because you can use the latest model for free, and can keep the session open, even as you run into usage limits. It's been fun and interesting, for a story in my mind for decades. I need to prompt a lot, and sometimes direct it to specific parts of the source material I uploaded, but I find it satisfying so far. Grok 3 is free but left me unsatisfied. You've written very positively on Claude Opus, but it costs, and I'm on a budget.
So far, I don't think my story has fallen into this uncanny valley, perhaps because of its unusual subject. It even suggested some unexpected phrases I found apt.
What other models do you recommend that might be good for creative writing?
What a great, short, informative post.
I have tried using Grok for some stories and one thing thats disables it is that Grok for each output always try to distill as much information as possible from previous pages into the next page. So Grok will always be telling the reader that they are in the "Blood Temple room" although the reader already knows this from 5 pages back.
Engfish
https://www.thoughtco.com/engfish-antiwriting-term-1690596
Antidote: free writing.
Chat's EngFish Scoreboard
Robert A. Heinlein – 8
Larry Correia – 10
Larry Niven – 18
John Scalzi – 22
Isaac Asimov – 25
George R. R. Martin – 28
N.K. Jemisin – 35
Kim Stanley Robinson – 55
Will promoting ai to write in a certain author’s style, or a mix of author’s styles, alleviate the problem? Such as if I prompted the ai to write a fantasy story in the style of Tolkien and Lewis.
Not even a little bit.
I tried this out and verified what you are saying. Would it be different if you built a local ai that only understood the two example authors? In other words, if I scanned all the works from two authors into the local ai would the outcome be different?
Jim, are you trying to prompt Vox? He has been accused of being an AI, so maybe you'll get an improved response.
I’ve noticed that AI when creating dialogue between two characters is often very abrupt, like it’s running down its checklist of bullet points that you provided in an outline. This makes it feel very unnatural as people tend to ease into uncomfortable topics.
Also, most surprising to me is how AI gets things wrong like a real person does. Sometimes events happen out of sequence even though I provide a timeline/outline of the sequence. And sometimes it “forgets” basic facts about the character/events.
I've noticed this. I've had to repeatedly point out the chronology of what it's writing, even though I provided it.
The idea that authors will become stupid using AI — I’m not convinced. It simply makes things faster, but the author still needs to screen, edit, and be prepared to rewrite absolutely everything.
Yeah, that's if you're conscientious. You're just saying, "**I** wouldn't become stupid."
The more the cost of something lowers, the more of it will be demanded.
Stupid is often paired with lazy, and if there's one thing that almost all of humanity is united in loving, it's the easy way out.
AI is like sports: it doesn't build character, it reveals it.
Biggest one is AI getting confused after a few rounds of editing. It can start to emphasize or add things completely irrelevant to the story's design.
Usually happens when discussing the work and potential futures. "No, AI, I do not want you to continue utilizing those hypotheticals tgat aren't going to be included in tje final draft." The solution is to start a new session to get it back on track.
I have yet to produce a "fully AI work" as the current process is to write the story and use AI for revising purposes while I make the adjustments by hand.
It does limit AI-isms as the AI is "airgapped" from the text.
Yeah, the constant attempts to foreshadow information you do not want disclosed is incredibly irritating.
i see what you did there