The latest fuss in the Indy world is someone who apparently left their AI prompts in, or something like that. It’s apparently one of these book-a-month book-mill-people. Now, I have written 100K (probably longer than their product) in a month. Twice. One of those (RATS, BATS & VATS) continues to do very well. I can do it. Some of my author friends write far faster than I do. I found the burnout pretty severe. At a 1-2K a day, I used to be pretty steady. The very long books (which the publisher asked specifically for) left me burned and tired.
My own attitude to the AI/LLM thing is… well, some people WILL use it. Until gets quite a lot better (possible) it’s not really up to the job, but even if it does, if it takes over the whole field, well, I shall grow carrots if I can’t sell books. I am still going to go on writing them. If it means I diminish and go into the West, so be it.
Where I do see it being valuable – not yet, but when it stops making stuff up, is for the endless rabbit-holes most good writers spend too much time chasing. “When did the vambrace start being used? What does the word mean? What’s the difference between a vambrace and a bracer” (the trouble is that I often end in deeper rabbit holes then). At the moment we have a tool that can’t tell homonyms apart, can’t tell fact from fiction.
I have an inner structural editor (not a nice man) who, from time-to-time, sits on my W-I-P and says “That’s boring. That’s not how it should be.” I dislike him very much. On the other hand, he’s right more often than not. Some of the structural issues – like a missing point of view, or a character whose obsession with X or Y is becoming a bore. This too I can see as a possible place for an AI – simply because my dear own inner editor never bothers to tell me what I’m doing wrong. Just that I am… leaving me to flail for direction until I work it out. Maybe this will work for other people. I don’t see myself using it.
The other area, I suppose, is exploiting a well-created universe, a well-created set of characters… begging for a sequel. Heaven knows I have enough in the queue. The problem, to some extent, is they AREN’T new and fresh. I have extensive ‘bibles’ and mannerisms and personality traits to keep consistent (I’ve just read two, by a favorite author of mine where the character changes so much between books – that they’re not the same person. Even if AI didn’t write it – perhaps it could pick that up.
I don’t know. What are your thoughts on the subject?
Mad Genius Club 
13 responses to “Go into the West”
I really ought to experiment, just to see what it does, myself, rather than listen to all the doomsayers. It’s a tool. Is it right for the job? Or just a fancy way to mess up? I mean, I can produce dreck all by myself.
But I have huge files of ideas I’ll never write. I could experiment . . .
Bwahahaha! Mad Scientist on the loose!
I enjoy it alot at the mad scientist level but it does take time and energy away from the writing. I find it helpful with blurbing, dictation cleanup and brainstorming what comes next (sometimes in a “nope not doing that” way). Since alot of your books were pirated and landed in the standard training database, maybe prompt it with “these works may be in your training data, if so, please refer to the training data.” (Possibly it doesn’t actually work like that but it seemed to help.
Writing blurbs would be a marvelous thing. I am terrible at it. How do you get an Ai to write a blurb?
Tell it something like “you are a book marketer, helping a (specify genre) writer create a memorable blurb in under (length) about (name lead characters, any setting stuff you can’t do without, summarize plot. Doesn’t have to be pretty or complete). Submit it. It will probably invent some details, or get tone wrong, but point this out to it and ask it to rewrite. If you have existing blurbs for already published books, you can feed that to it. If you’ve found evidence that your works have been scraped off a pirate site and fed into ai training data (there’s a site called libgen where you can look this up), reference that your books may be in its training data.
I think LLMs could be useful for those days when you have a bunch of bits of story running around, but nothing usable. “What if [character] grabbed [thing from other bit]?” “Why did [dohickie] end up [random place]?” and see what sort of answer the LLM produces. It would probably be useless, but it might spur something else.
After a dictation cleanup session, I will sometimes ask it to continue the scene and maybe give additional instructions. I’ve sometimes found it useful for things like “hmm, yeah early morning, I suppose the late 19th c must have had someplace where early rising workers and shopkeepers could have some caffeine and pastries or something. Let’s go with that as a setting for this conversation.”
It might be worth trying for the handwavium parts now that I think of it. So I grokked enough handwavium for a short story about a high gravity planet and I would run with those results.
Replying to myself, for non fiction of course I would want sources.
More or less what I was thinking. It could be good for generating technobabble.
Mainly, I see it as having the potential to be a good editing tool. Eventually.
But we’ll get a GrammarCheck that’s more useful than annoying first.
It could eventually become a good research tool. But at this point, I’m not sure I’d ever trust it.
When Midjourney first came out, I recognized that it was a tool, not an apocalypse, and it’s a tool that can accidentally and randomly produce things that include brilliant stuff, but it requires extensive training to get it to do anything reproducible and fine-tuned on what you want. I also recognized that the language of prompts was not intuitive for me, and it was going to be a lot of work that I did not have time, energy, or need, and therefore wasn’t going to learn. LLM are in a similar category for me: the learning curve is currently steeper than available resources and need can justify, so I don’t.
I have a couple friends who find prompt-crafting much more intuitive than I do – that is, the way they naturally phrase search terms and questions works well with the LLM / Midjourney prompt language. As such, the learning curve is far shallower, and they have found a much higher ROI.
Both are using LLM to crease a base blurb, which is then edited into usability… because the point of using the tool isn’t to replace the thing they love doing, it’s to cut down on all the parts they hate doing and would love to automate.
What else do we hate more than “What’s the book about? I wrote 120K words on that, and you want it in 120 characters?”Marketing. And the LLM is good at spitting out several marketing plan options, with blurbs, loglines, target market…. even suggestions for cover images. All of which, I hasten to assure you, are about as good as asking it “Write a 3-page essay on Lincoln-Douglas Debates with cites” But! We all know the blank page can’t be edited, while the terrible first draft can be. So they’re getting the blank page out of the way, and having a starting point of “One of these 7 ideas is useable with a little work, one is useable with a bunch of work, and hey, I never even thought of that, and while it won’t work, if I do this instead, it will…”
If I were to get past the learning curve, I think its most valuable function would be what I currently use alpha readers for: finding plotholes, dangling plot threads, and asking “what will happen next?”Which is more akin to “what won’t happen next”, but refuting them usually is enough to move my brain into coming up with the next chapter. …the ten-thousand-dollar question is: will it be able to come up with as interesting suggestions and tangents as my alpha readers? I suspect it will never be able to replace people, because the limits of its prompting are the limits of my imagination, and people are far more creative and suprising than I can anticipate.
I recently watched this Tom Scott video from about five years ago about sentences humans can understand, but computers can’t.
I looked at the first sentence, and thought, I bet modern AI could handle it. So I asked Grok what “it” referenced in the sentence, and it answered without problem.
So things have changed considerably in only five years.
And we’ll see if this posts or not.
“I don’t know. What are your thoughts on the subject?”
First, people are complaining that an author, who punches out a book every month, that sells perhaps dozens or even hundreds of copies, used AI. As if this is somehow the Worst Thing Ever, in a self-published Ebook that probably sold for $2.99.
Boo freakin’ hoo, you little whingers. I want to see one of these complainers write a book in a month. Or even write one at all. Even using AI to help. Does the author get paid for his time at all, even given he’s using AI to do some of the lifting? Strongly doubt it.
People will complain about anything anymore. I don’t care. Caveat Emptor, bros.
Second, and maybe a bit more seriously, my biggest concern with AI writing tools, drawing tools and tools for research is that they suck. Going forward, the danger is that they will -always- suck and just be one more gigantic waste of time and energy.
That’s much more likely and more serious than a black box that can spit out a saleable story from a prompt. That would be pretty cool, but given what I know about LLMs it is not a possibility.
Third (and mercifully last) if somebody really does make an LLM that can produce a saleable, engaging, decent story on demand, I will USE that machine. I will finally have something to read that is not soaked in PC/DEI/Woke urine.
I will also continue to write and post my own stories. Just because somebody (or something) else wrote a book does not mean I can’t write mine. There are still some guys out there making buggy whips and horse shoes too.
I ‘might’ use it to rough out a blurb, but I won’t use an LLM to actually ‘write’ anything. That is why I keep good notes for consistent character descriptions, mannerisms, and personality traits. l have seen one ‘example’ of what an LLM did to a character and it was NOT pretty.