[sieve]

> inconsistency across various parts of the story, items that are introduced, questions that go unanswered etc.

This is a workflow problem.

What the LLM normally needs:

- very basic outline: what the story is about

- basic cast of characters: like fiction/plays of the 18-1900s

Then you go chapter-by-chapter. But the most critical thing is that you MUST provide the chapter beats. From start to end. Location, environment, events, any critical conversations. For a chapter that is 2-3K words, this chapter outline often becomes 2-400 words or more. Because you have the story in your head, you can handle the stuff that the LLM is likely to screw up.

There are ways to build a system where the LLM can keep track of many things, but then that becomes programming, which is just as boring.

[illwrks]

I won't go into too much detail, but I have a plot outline, development bible, story mountain, character outlines, style guide and more. I have four agents running on various tasks; writing, error checking, consistency and critique. They can generate a good story but there are holes and issues so it's far from perfect and I wouldn't give it away for free, never mind put a price on it.

To add to this, in my 9-5 I work as a designer. Colleagues in other parts of the business sometimes come to us with stuff they have created using AI tools. To a layman they look really good, but if you know what to look for you can spot the issues.

At this point in time my opinion is that any AI generated thing is great at making fools of the unaware, or those who are not looking at something critically. It can create something that looks like a duck, walks like a duck, quacks like a duck, but on closer inspection it's not a duck.

[sieve]

I tried the programmer's method. Automating stuff. Then I realized I prefer the wood carver's method. So I take it slow.

I think and write the outline for a chapter. Then I have the LLM produce it. I read the draft. Tweak things. Reroll. Take the best from multiple rolls and merge them to produce a organic whole. Manually patch the minor differences in pacing and tone.

Rinse and repeat.

It takes 100-250 hours across 3-4 months to produce a story. But I drop 99.9% of the stuff within the first day or two because the plot or characters do not work.

I use the web front end directly. Let's me switch between any model including Gemma 4 running locally.

[illwrks]

I think that's probably the best approach right now.

[grey-area]

No, this is a problem with the nature of LLMs and the text they generate.

[sieve]

I do not expect the LLMs to do everything. They are assistants. You still bear the responsibility of stitching everything together into a cohesive whole. Make changes as required.

That is where expectations differ I guess.

[grey-area]

But they generate objectively bad text. Why do you need an assistant to do that?

Perhaps for some uses (say a summary of some business document, or generating code which is then discarded or edited) that's acceptable. For literature, do you really feel they are useful tools? So many changes would be required IMO that you're better off just writing the story you want to tell.