Adding AI tools to Scrivener

What if the human cut up 1000 novels, then pulled out all the pieces from novels written by Stephen King, then pulled out all the pieces from novels about rabid St. Bernards, then assembled those pieces into something that repeated large sections of Cujo verbatim?

Prolific authors – like King – and large organizations – like the New York Times – are overrepresented in the data sets. As a result, there’s quite a bit of evidence that verbatim excerpts from those sources are overrepresented in the output as well.

That’s a problem for the LLM companies, but it’s even more of a problem for authors using LLMs to “write” (and their publishers). If ChatGPT spits out a pastiche of Stephen King’s work and I put my name on it, his lawyers are going to come after me and my publisher, not (just) OpenAI. (And probably OpenAI’s terms of service immunize them from any responsibility for the output.)

6 Likes

The 2026 internet is a very different place – both better and worse – than the 2001 internet.

Whatever follows LLMs won’t do the same things or work the same way.

2 Likes

Again…

2 Likes

I’m going to spend my time doing more productive things. Probably with help from AI.

FYI, I still write all my pages myself, because that’s the fun part of writing. But wow, is AI great at doing research and digging around in my own notes.

It’s too bad Scrivener won’t have an MCP so LLMs could automate things.

Very good advice. Me too.

I’ve used it a couple of times, and I work in a company where people use it extensively for research, summarising positions, reviewing documents and notes, and preparing first drafts of things… and in my experience, AI has consistently produced sub-standard work (as in “if-a-human-did-this-they’d-be-getting-talked-to”).

As an aside, I expressly asked AI for the robot to have 7 arms holding 7 pieces of paper. I hope your research assistant is more reliable!

3 Likes

The unattached arm is nice.
And hey… 4 on one side and 3 on the other, just like your standard spider.

[EDIT] Took me a while, but I finally spotted the 8th one.

What I like a little less :

(..and twice, rather than once.)

I asked Gemini (for the irony of it) :

You turn stupid, in other words.
(Doesn’t apply to those already stupid to begin with, though. Or does it?..)

1 Like

I did like the highly useful reminder it put on the notice board, though…

2 Likes

I’ve talked about this before, I think, but I’ve personally observed a significant and material drop off in analytical skills and critical thinking amongst 100% of staff who use AI regularly[1]… and of course this also means that there is a group of people out there who will never develop those skills.

That’s not to say that this is necessarily a big deal. Calculators mean that my mental arithmetic skills have atrophied enormously but I’m kind of okay with that because I have calculators. And my oil painting skills never developed at all because I have cameras. I PERSONALLY think the loss of analytical skills and critical thinking is a big deal, but I understand why others might hold a differing opinion.


  1. and this is NOT a small sample size anymore ↩︎

4 Likes

Any form of automation, whether by LLMs or human-written scripts, would first require the development of an API.

1 Like

The site about fallacies that you reference is very good. (It’s the best part of this conversation, IMHO):nerd_face:

1 Like

Well then it’s a good thing the courts never get it wrong or rule in favor of the more powerful and wealthy at the detriment of the poor, marginalized, or underepresented. :eyes:

9 Likes

I’m very much on the side of @pigfender, and anyone else who thinks ‘ai’ , really of any kind, hasn’t a place in Scrivener or any app intending as it does.

I would never let such near where it could provide text.

However, besides learning in some depth how it can work, at least with the kind of ‘thinking’ that Gemini Pro now has, I’ve experimenting in using it some ways which I think do amplify at least the sort of novels I’m working out how to write.

What I’ve learned, I think would apply to those in pigfenders office also, potentially at least.

  • if you just ask it do do things for you, you are cooked; all the lines he suggests, I would also feel
  • if you just ask it for answers, this is also going to get rote results, even as cleverness has risen over hallucination and plain repetition of all the misinformation out there on the web.
  • all warnings about how it will tell you how brilliant you are, and otherwise be soft in ‘alginment’, one must surely treat as forewarning, if you can learn to modify some of that

However.

What I’ve found that opens doors well beyond this, is to hold real conversations. Not usually along lines it is also prepared to suggest. Rather, the kind of sharp, calling on all points ones you’d have with a best educated, intentional partner for such; a sharp but experienced graduate student assistant may be a good model.

You want to do two things:

  • Interrogate whatever it tells you, ask it to elaborate, explain, clarify, substantiate (asking for links will often do the last job).
  • And this is the obvious, which will open many doors, just as if you were doing if for yourself at the library.
  • what goes very far beyond is to pay very aware attention to those little suggestions which come into your own mind, as they willi in any conversation….
  • don’t hold them back, no need to be polite here, and the converstant will never bite (see alignment, as they call it).
  • Say what occurs to you, just as it does. Take the conversation on a tangent, as far away as your sight has seen. For this is exactly what the machinery cannot do.
  • Once you give it the chance, though, what it can do is go look very much faster and farther than you could, to dig up on that insight or tangent, and bring you back that information you’d be so pleased if the grad assistant would.

The key is that you are very much better at ‘seeing’ that original connection in its sense of potential meaning. And then what the machine is so good at, access, gets good signt of it immediately returned to you.

Which likely causes another interrogation to ensue, a situation or a scene clarified, all the things that you manage to do when you have traced such a moment of sight down yourself. But much as if the library delivered the papers and the books right to you – and your conversation continuing, so you could explore it with the person of knowledge, but not your field of judgement.

I find this kind of in its way Socratic discovery very effective, for things like grounding experience in settings and histories, for extrapolating patterns of social or political nature which will mesh with the personalities and attentions, the intents of characters, and trajectories of their surround.

The ai is not writing a thing for me, not any word. But it is letting me hone my attentions, my backgroundings, and even the fluency of typing I can transfer directly to comforts in elaborating what happens among the constituents of a story.

Early days on a very deep story, to say the least, for work here. Quite an amount of experience now, doing what I describe, feeling the effects, which are certainly deadening down nothing; quite the opposite.

Would it fit an office scenario, @pigfender’s friends in particular? That would of course depend, on persons, on what they believe they are doing, and perhaps most of all on how all that doing might be able and be permitted to expand. By having better stories, I might suggest.

But again, that’s very individual. And miles away from any temptations or mechanistics we would be quite correct to instinctively and wholly resist, from coming into any proximity of Scrivener.

You don’t invite the human friend to sit on top of your writing table, either, do you….

2 Likes

[above suitably edited, I hope, to be more readable on the day after elections…]