Review

I've been worldbuilding a unique setting in my head and on paper for maybe a year now, and for quite a bit longer than that I have wanted to start publishing fiction. I've never "found the time" to act on these ambitions, of course, but until the advent of LLMs I always thought I would have the opportunity to work on it later, after this or that startup or video game or work thing.

I might still do that, but I'd be lying if I didn't say that part of the appeal of writing fiction anything at all was to receive credit from a very small group of readers for a piece of content that they liked and found novel or creative. It seems to me like the possibility of that is either fading or has already faded; the one big barrier that I thought was preventing LLMs from writing longform content, context size limits, has been blown away. GPT-4 now has a context size well above that of most short stories, more than enough to summarize the essential beats of a novel down from ~200 or so pages, and my guess is that future people will assume that most of the words and story details are being written out of a machine. Perhaps the basic idea for a story will be the domain of humans, but then people will be left wondering exactly how much of that idea was the product of the author, and rhetorical giftedness will become commoditized. That in mind, should I get started now, or is this worrying too much, or is it already basically too late?

New Answer
New Comment

5 Answers sorted by

awg

121

I too am an aspiring writer and have been for the past decade or so. I've written a novel-length story and recently I've written about 50k words worth of short stories in anticipation of trying to get published sometime soon. In the meantime I'm a software engineer using generative AI in my work every day and I'm keeping up with the dizzying pace of changes that have been coming. So, I feel you. I actually wrote a story about this to help me process all of it a little, heh.

Anyway, I suppose there are many ways to look at what's coming, but here's how I've been trying to look at it:

  1. No AI will be able to take the joy of creation away from you. Remember that first and foremost you worldbuild and write stories because doing so brings you satisfaction and fulfillment on its own. That will never change, no matter how many AI-written masterpieces might be getting churned out in the future. It can be easy to get lost conflating the act of creating with the praise one might get for one's creations–but try not to. Try to hold the simple joy you have in the act of creation itself. It is good and meaningful in its own right.
  2. (I believe) Humans will always crave human-made things. Reading a book from your favorite author isn't just you blithely ingesting the story they've written. It's a connection between you and them and between you and their world. We read novels written by human authors not just because the stories are entertaining, but because we crave to understand other humans. So, I think even if AI are consistently churning out new and wonderful worlds and stories, there will always be a market for what humans can uniquely create, as long as there are humans uniquely creating. (And I think there will always be humans uniquely creating, because of point 1.)

The printing press revolutionized the way humans wrote and shared information. I think generative AI is transforming the way humans write and share information similarly. But I claim that no revolution in technology will dampen the inherit desire in humans to share and relate with one another directly, genuinely, and emphatically.

The novel is as popular now as its ever been. More people have the opportunity to read now than there has ever been. There's no reason to be discouraged from participating in the wonderful act of creating and sharing, even (or especially) now.

Humans will always crave human-made things.

If true, then this creates an incentive for falsely presenting as human-made that which is not. If doing so is easier than actually making the sort of thing in question, then the obvious outcome is that the market for such things will be dominated by things that were created by AI but are presented as “human-made”.

[-]awg10

Following up on this with something from Wired

A full 15 years ago, a team of researchers from University College London and the University of Copenhagen put people into an fMRI machine and showed them a series of abstract images. They told them the images were either made by a human or by a computer. A clear winner emerged. People not only claimed to prefer the (identical) human-made pictures, their brains’ pleasure centers actually lit up more brightly. What the researchers didn’t anticipate, but which is likely to happen, is that th

... (read more)

eggsmediumrare

85

I don't want to read a story someone coaxed out of an LLM. I suspect I will someday and won't even notice, but if I was then made aware, I'd be mad about it. I wake up every day and write 1000 words on a story I'm pretty sure no one will ever read. It probably seems like a waste of time, but I enjoy it, so it's not.

[-]awg21

Thank you. You said in one simple paragraph what I tried to say in my whole, long answer!

Ben

51

I love worldbuilding and writing short stories, and I have thought a little along these same lines.

I think that the LLMs will help with people writing stories faster. Perhaps in the near future given the text so far and the plan some scenes or sections will be drafted by a LLM that will then be edited or rejected by the human author. Other tools will probably exist, maybe a LLM tool that can proof read stories and do more than just spell-check, but also raise other issues. (Ranging from "this character is blonde in chapter 4 but a redhead in chapter 7", through to "The beginning is quite slow." and "Character X is really central to the plot, but seems to be poorly defined.")

I suspect that you might be exactly the sort of person to benefit most from this. I hang out in fantasy writing groups, their are two broad categories. People who like the actual process of wiring prose, who tend to write relatively "safe" (standard) settings and plots, with smooth text; and people who enjoy building intricate and unique worlds, but as soon as they start actually writing text they are bored to tears after two pages and by then they have an anxiety complex about where comers go in relation to quote marks and ten tabs open telling them esoteric stuff they never knew like "em-dashes are better than ellipses" or "their are two types of third person narrator (either is fine but never switch!)". They find themselves thinking "urgh, I just need a scene where they meet, establish these facts and move on. But it needs to look like story text, so I need to pad it out with a description of what the nondescript office looks like - why am I doing that again?". The latter group (which it sounds like you might belong to*) rarely finish any projects, but hopefully LLMs will change that.

But I don't think that stories "cranked out by machine" without any kind of human intervention are ever going to be a major thing. The cost of introducing a "quality control" human who can decide which of the ten-million stories the program cranks out are actually worth publishing is small. (If you publish all ten-million then how many do you expect a typical example to sell? Printing economies of scale want you to pick a smaller set to publish.).

From my playing around, current LLMs are OK at prose, but are weak at plot and structure. 

I also think their is an exciting new age coming in terms of writing styles. Maybe in  ~10-15 years their will be a recognised "LLM style" of writing that it will be fashionable to deviate from in big ways. Maybe intentionally poor grammar, more likely going off in plot directions it tends to not do (I find chat GPT hates conflict in its stories and largely refuses to include it**). A bit like how painters went all cubist after photography came along. 

Also, text adventures are probably going to make a comeback, if they haven't already. Something I think would be fun is if me (or you) could give the LLM a setting, and then I could roleplay a character in that setting with text adventure. It gives another way to share a world you have built.

 

*  I bet that you have a vague idea of what the biggest industries in some fictional country in your setting are, but you have no idea which type of third person you are writing in.

** I asked it for a story where J. Edgar Hoover met Del Boy. Started reasonably, Del was in Washington and had a stall selling all kinds of obviously stolen goods. Then Edgar turns up, they become great pals and Del is recruited into the FBI. Given how much else it was getting right the lack of conflict between these characters stands out as bizarre.

50

Conversely why not adapt. Be the first person to finish a GOOD million word fanfic with the individual effort of writing a 100k word fic? People don't adapt to the new tools instantly and some of the new outputs coming from the latest hacked version of stable diffusion blended with other models are astonishing, like as in "some of the best computer generated images ever made".

I mean millions of people before you have written a story. Hundreds have written a shitty story with the help of AI. GPT-4 with careful guidance might be able to write a story that doesn't suck.

DirectedEvolution

2-4

Good artists borrow, great artists steal, God just uses ChatGPT.

[-]awg41

This is a snarky joke answer with little actual content and I'm kind of upset it has the most upvotes for a question of this level of earnestness.

4Said Achmiz
Perhaps other people perceive more “actual content” in this answer than you do? (I, for one, read it as a “ha ha only serious” sort of comment.)
4DirectedEvolution
I agree, it’s really frustrating and sad to see how badly the standards for commenting on LessWrong have declined.