On writing with AI: The Devil is in my foyer, do I invite him in?

  • Thread starter Thread starter DaveC426913
  • Start date Start date
AI Thread Summary
The discussion explores the use of AI in writing, particularly in the context of researching and developing characters for a short story set in 1850s New Orleans. The author reflects on the benefits of AI-generated suggestions for names, professions, and character traits, while grappling with the concern that relying on AI may compromise the originality of their story. They emphasize the importance of distinguishing between using AI for factual research and for generating creative content, suggesting that the latter risks diluting the author's voice. The potential of AI as a tool for editing and ensuring consistency in character dialogue is also considered. Ultimately, the conversation highlights the need for clear boundaries in AI usage to maintain the integrity of the writer's work.
DaveC426913
Gold Member
Messages
23,886
Reaction score
7,916
Man, when the Devil comes a-knockin', he does not smell of sulphur and brimstone, he smells of sweet perfume and roses...

Doing a bit of research for a short story. All I wanted was some examples of professions typical for jurors in a trial of the time and location of the story (1850s New Orleans - of which I am not very familiar, so I have my work cut out for me).

ChatGPT delivered that very nicely, giving me a list of a dozen typical professions for the type of man that was eligible for jury duty (and also told me women were not eligible) allowing me to pick and choose and start developing some characters. That's just research, right?


And then ... in little text at the bottom, it said - so demurely - 'if this is for fiction, let me know, and I can add some examples, along with some typical names - even some and personality traits and quirks'.

Oof. So I could do all this research and development on my own, one-by-one, for each of the dozen characters, or I could just hit 'Yes' and the clanker could give all this to me wrapped up in a bow.

I mean, look at this wealth of story-related content: french names (Monsieur LeClerc, Jereemia Boone), cultures (Creole), professions (shipwright, planter, lawyer), motives (blue collar upholding loyalty and labour, businessman who wants lawbreakers to be held accountable for disruptons, lawyer climbing the ranks).

How do I say no? (It is even possible to say no? Having read it, can I forget what I read? Can I unring this bell?)

I mean, I've aleady cracked the seal on the AI genie bottle. I already have to put in the disclaimer 'AI was used to do some of the background research that formed the basis of this story'.

How much worse could it be if I changed that to say 'AI was used to make suggestions for some of the content of this story'. Is that so bad? Is this how it starts?
 
Physics news on Phys.org
If I were writing that story, I'd book a few days in New Orleans and walk around the old cemeteries, maybe take in a few plantations along the river.
 
  • Agree
Likes symbolipoint
I think there is a difference between using the AI as an advanced search machine where you ask it for specific factual data that you can then verify (to weed out confabulations) before actual use, or ask it to generate any kind of fictional content for your story. If you go with the latter then the story to some degree is no longer your story.

Let's say you seek inspiration for some background material in the story. Traditionally you would research the area and time reading up relevant material and texts while (hopefully) keeping track on what is fact and what is fiction so that you do not accidentally use another authors fiction directly as your own fiction or, worse, use it as facts. If this is something you would still like to adhere to but want to use a chatbot to save some time and effort, then one way would be to search for fictional work that intersects with your story settings and then go read parts of such work for inspiration just as you would without a chatbot.

If nothing else, at least you have correctly identified that a modern LLM offers a drug of the most addictive kind and that establishing clear limits for your use of it up front is a very good idea.
 
One thing I thought LLMs might be quite good for is editing. Is your Floridian trailer-trash character suddenly talking like an educated New Yorker for a couple of pages? Are there authorial catchphrases turning up in every character's speech? It might also be able to catch stuff like one of your characters knowing something in chapter 20 that he isn't told until the big reveal in chapter 30.
 
Filip Larsen said:
If nothing else, at least you have correctly identified that a modern LLM offers a drug of the most addictive kind...
That's very much what it is, isn't it?

I think there is a difference between using the AI as an advanced search machine where you ask it for specific factual data that you can then verify (to weed out confabulations) before actual use, or ask it to generate any kind of fictional content for your story. If you go with the latter then the story to some degree is no longer your story.
This is what I am telling myself.

The chatbot produced a list of a dozen plausible names for the time and culture, with a smattering of back story and a quirk or two thrown in. If I included those character descriptions as-is, not only would it begin eroding at the concept of "my" story, but it might even be a less well-written story, since those elements do not advance the story.

But I guess there's nothing wrong with taking inspiration from its suggestions, separating the wheat that develops the story from the chaff that does not.

I had not (yet) thought of adding a Creole aspect to the story's cast of characters, so writing that in will definitely enrich the story. On the other hand, it suggested a lawyer character that has a motive to climb the ladder to position of judge. That doesn't further my story so I ignore it.

OK, that's a pretty clear delineation. That, and making sure every word in my story is placed there by me.
 
Last edited:
Ibix said:
One thing I thought LLMs might be quite good for is editing. Is your Floridian trailer-trash character suddenly talking like an educated New Yorker for a couple of pages? Are there authorial catchphrases turning up in every character's speech? It might also be able to catch stuff like one of your characters knowing something in chapter 20 that he isn't told until the big reveal in chapter 30.
Can it do this? I've got a 14,000 word (38 page) short story in its final draft, having gone past several amateur colleagues. It'd be useful to have the input of a chatbot, but can it handle something that size?
 
So I invited the Devil in. And when I wasn't looking, he peed on my furniture...

1759103646777.webp


In all my stories—collectively, 20,000+ words—I have completed the conversion of every parenthetical—of which I use a lot—to em-dashes.

I did not know it was a hallmark of AI. And I don't know how to unring this bell. Do I ... replace them with commas? Brackets?
 
Are you asking how to avoid writing your AI inspired story so it doesn't look like its completely written by AI?

Or are you more asking how to make good fiction prose where the sentences flow more naturally for the reader without heavy use of commas, dashes or (horror!) parenthesis?

If it is the latter I would be interested in some good pointers too :biggrin:
 
Filip Larsen said:
Are you asking how to avoid writing your AI inspired story so it doesn't look like its completely written by AI?

Or are you more asking how to make good fiction prose where the sentences flow more naturally for the reader without heavy use of commas, dashes or (horror!) parenthesis?

If it is the latter I would be interested in some good pointers too :biggrin:
Story is entirely mine, 100%. Story ideaxis mine,100%. I only discovered AI could read and edit stories a few weeks ago.

But AI makes a good proof reader. It recommended em-dashes—which have always confused me as to their application—so I changed them. Only to find out that they're a hallmark of AI.

Brackets look too informal in serious fiction and hyphens are wrong.
 
Last edited:
  • #10
DaveC426913 said:
Story is entirely mine, 100%. Story ideaxis mine,100%. I only discovered AI could read and edit stories a few weeks ago.
Sorry, after your OP I couldn't help tease you a bit there :wink:

DaveC426913 said:
But AI makes a good proof reader. It recommended em-dashes—which have always confused me as to their application—so I changed them. Only to find out that they're a hallmark of AI.

Brackets look too informal in serious fiction and hyphens are wrong.
And I guess foot notes aren't going to work well either.

I seem to recall some authors prefer to structure "inserted sentences" as a triplet of sentences instead. The trick is then to make the second and third (or more) sentence start in a way so that the whole thing works as if using dashes. I think that works nicely if there is a context switch, e.g. "Something happens. Inner thoughts. Then something more happens" as opposed to "Something happens—inner thoughts—something more happens". Also works well for a train of thoughts that can jitters around as thoughts tend to do, interspersed with small sections of action.

But what do I know, I only write technical non-fiction prose.
 
  • #11
Filip Larsen said:
Sorry, after your OP I couldn't help tease you a bit there :wink:
:-p
Filip Larsen said:
I seem to recall some authors prefer to structure "inserted sentences" as a triplet of sentences instead. The trick is then to make the second and third (or more) sentence start in a way so that the whole thing works as if using dashes. I think that works nicely if there is a context switch, e.g. "Something happens. Inner thoughts. Then something more happens" as opposed to "Something happens—inner thoughts—something more happens". Also works well for a train of thoughts that can jitters around as thoughts tend to do, interspersed with small sections of action.
It wasn’t until he was almost upon it that he saw the bulk of the vessel moored there. Its silhouette against the moonless sky exposed—to the expert eye of the Shipwright—the unmistakable twin smoke-stacks of an ocean-going packet steamer.
 
  • #12
DaveC426913 said:
Its silhouette against the moonless sky exposed—to the expert eye of the Shipwright—the unmistakable twin smoke-stacks of an ocean-going packet steamer.
I would think for this example at least it should be possible to rephrase without too much change in meaning or flow?

I mean, structurally something like:
To the expert eye of the Shipwright its silhouette against the moonless sky exposed the unmistakable twin smoke-stacks of an ocean-going packet steamer.
 
  • #13
Filip Larsen said:
I would think for this example at least it should be possible to rephrase without too much change in meaning or flow?

I mean, structurally something like:
To the expert eye of the Shipwright its silhouette against the moonless sky exposed the unmistakable twin smoke-stacks of an ocean-going packet steamer.
Yeah. I do prefer the pacing of the em-dashes - that is how I normally write. Yours is a bit clunky to my ear.
I just didn't know my personal style was also the steyle of chatbots.
 
  • #14
I was also thinking about breaking it up into more sentences.
DaveC426913 said:
In all my stories—collectively, 20,000+ words—I have completed the conversion of every parenthetical—of which I use a lot—to em-dashes.
editor said:
My stories altogether run to 20,000+ words. I originally made heavy use of parentheticals, but recently my AI editor suggested I use em-dashes instead. I have completed the conversion.

My personal belief is that either parentheticals or em-dashes capture a more stream-of-thought style, where the author (speaker) is writing (speaking) off-the-cuff. Making it up as he goes along rather than keeping silent until it is a finished piece.
 

Similar threads

Back
Top