ChatGPT, the future of authors, and how artificial intelligence can help you write.
*This blog post is available as a YouTube video if you’d prefer to listen instead.
Prologue: What is ChatGPT?
When I was a kid, I wished for a robot to do chores for me. Now I have a robot to do my work for me, but is this as cool as it sounds, or should we be worried?
This is Legends of the Shadow Woods, and though I am publishing it I didn’t really write it. Or did I? I used ChatGPT to write this Novella, and it’s available as an ebook on Amazon for 99 cents.
Legends of the Shadow Woods
Embark on a journey through the mystical Shadow Woods with Legends of the Shadow Woods, a collection of tales based on Greek mythology. Follow the hero Ajax as he battles sorceresses and monsters, uncovers ancient artifacts, and discovers his true destiny. These action-packed and thrilling stories will keep you on the edge of your seat as you delve deeper into the rich world of the Shadow Woods.
AI wrote that blurb too.
Along with using AI to write the book, I used a different AI to make the cover.
I’m sure everybody has heard of ChatGPT by now, but if you’re not familiar:
ChatGPT is a computer program that can understand and respond to written language like a human it can generate text that is coherent and contextually relevant by making use of the input it receives and by learning from the examples of text it has been trained on.
And guess what? ChatGPT wrote that part too.
I’ve seen people using ChatGPT for all kinds of things, from writing essays and resumes to creating poetry and songs. In early January, I saw a video of someone using ChatGPT to write a children’s picture book, and got the idea that “hey I can do that too.” So I did. But instead of a children’s book, I wrote a fantasy novella.
Initially, I wasn’t planning on sharing this, because I just wanted to experiment with the technology. But as I started working on it, I realized I had so many notes that I thought somebody else might find it interesting.
Chapter 1: Generating the Story
Though I started working on Shadow Woods on January 9th, I was using ChatGPT for a few days before that learning how it works and playing around with it.
I tried not to influence it too much with story direction. When creating a world for the stories, I straight up asked it “What kind of fantasy world should I write about?” And here are the responses it gave me:
- A world with multiple intelligent species living together such as humans, elves, dwarfs, and dragons.
- A world with a unique Magic system such as one where magic is fueled by emotions or one where only certain people are born with the ability to use magic.
- A world that is inspired by a particular historical time period or culture such as ancient Egypt or fetal Japan.
- A world that is entirely made up of one biome such as a world entirely covered in dense rainforest or a world made up of floating islands in the sky.
This is a great example of how most of my prompts and answers went. I would ask ChatGPT for some options, and I’d choose from the options it provided.
I went with a fantasy world based on Greek mythology because I thought it would give the stories the best chance of succeeding. I’ll get more into the limits later, but I thought Greek mythology would work well because ChatGPT already has a large base of information to work off of. Another reason is that the style of the text it produces is kind of… vague? It reminded me of how mythology stories are told, so I thought it would be a good fit.
I think in its current state, it would be difficult to have ChatGPT write a contemporary novel, but I’m always up for a challenge.
I figured a collection of short stories would be easier to manage because of ChatGPT’s format. It makes it difficult to refer to earlier details, especially when regenerating prompts multiple times, there’s a lot of content to sift through.
The first story in the collection is called “The Golden Feather of Friendship” and the A.I. actually named every story, chapter, character, and place, and titled the collection itself.
Creating the first story was mostly trial and error, as it surprisingly took a while to learn what prompts will get the desired results. I initially worked in chunks of multiple paragraphs, but this complicated things. It was much easier to work in pieces of three paragraphs or less. It was very much “seat of the pants” as far as characters went, but I did have it create a basic outline because I’m an outliner and that’s my nature.
So I would give it one or two of the story points it generated, and ask it to create a chapter outline. Then I would ask it to make adjustments when it came up with something that didn’t fit, and I’d save the outline for future use. After that, it was a repetitive process of prompting for a few paragraphs, copying and pasting, regenerating, making adjustments, and prompting some more. And eventually, I had a very rough draft of a short story.
Chapter 2: Editing the Beast
I started generating on January 9th, and I was done generating, editing, and proofreading the book by January 21st. The final product is just shy of 25,000 words. I worked on one story at a time, so I don’t have a starting word count, but my estimate is there were 35,000 to 40,000 words of generated content before any edits.
My goal with editing wasn’t to bring it up to my standard, or to use my style, but to work with what I had. I wanted to produce something readable. That’s it. Not good, not awesome, just readable. And I think for the most part I attained that.
I really tried to preserve the generated content wherever possible, as this is kind of a showcase of what this AI can do. Editing it wasn’t too dissimilar from my typical editing process for my own novels. First a rough edit, then a more detailed pass using editing software, and finally a proofread.
The rough edit took the longest. There was so much generated content to cut. There are two main reasons for this: Firstly, I would regenerate sections of the text multiple times and then Frankenstein different versions together. Editing these into a semi-coherent chapter was pretty time-consuming. Secondly, ChatGPT tends to repeat itself A LOT. For a while, at the end of every single paragraph, it would summarize the character’s goal, so most paragraphs ended with a phrase like “And with that Ajax knew he had to defeat the dragon once and for all.”
I didn’t put as much time into editing this project as I do other books, mostly because the more time I spent editing, the more I wanted to change. I had to work through it carefully to preserve it without overdoing my edits. There were things I really wanted to change, but I left them because they’re kind of… part of it. There are out-of-place sentences and lines of dialogue that are out of character or totally unnecessary, but I left them in whenever they weren’t intrusive.
I changed a lot while I was editing. In some places, I took the editing further than I originally intended, but at times it was easier to write a few lines of transition or fix the dialogue myself instead of having ChatGPT generate something better. But overall, AI wrote most of the content.
If I look at the time I spent on this project, it’s about equally split between generating and editing time, which surprised me. I expected generating to take longer, but with all the regenerating I did, it makes sense.
And obviously, I didn’t do any professional editing, and I didn’t have any Beta Readers for this thing. Plus I only spent a couple of hours proofreading it.
Chapter 3: ChatGPT’s Limitations
While ChatGPT can do a lot, it still has some limitations.
Firstly, OpenAI has a Content Policy that says you can’t use the program to generate certain types of content. Obviously, they need this policy for a multitude of reasons, but the limit to violence makes generating action fantasy stories a little challenging. Once while I was generating content, I got a warning that said “This content may violate our content policy.”
My prompt was not specifically asking for violent content–in the context, I understand why it went there, but also it’s not what I asked for. I didn’t use any of the content generated from that prompt, just in case.
A much bigger limit for me is its clumsy sentence structure. I think this limit will be improved rapidly, but for now, it makes generating stories kind of difficult and requires a lot of editing. I expected the generated text to have better grammar, flow, and structure, but a lot of what it generates is wordy, repetitive, and overall clumsy.
But I guess it is good to know real robots struggle with clumsy writing as much as I do.
So on January 9th, ChatGPT had an update with improvements to the model for improvements across a broad range of subjects. This was the day I started generating these stories, but I was using ChatGPT for a while before that, and I think I saw improvements, but I don’t have enough data to say for sure.
Another limit or problem for me is the repeated phrases. On the OpenAI website, you can find some other known limitations, but I ran into this one the most. On their website, they say “the model is often excessively verbose and over uses certain phrases” I found this to be true, and it took a lot of extra editing time to cut out the “excessively verbose” content.
One more funny problem is that sometimes the AI forgets what it already wrote. It sometimes changes details along the way, like the color of a character’s eyes. This wasn’t something I expected at all, especially considering how complex of concepts it can keep track of at other times.
Chapter 4: What About Authors?
What does this mean for the future of authors?
Honestly, I don’t know.
I’ve been watching programs like DallE2 and Mid-journey freak out artists for months. Suddenly anyone with a few bucks for generator credits can create artwork with a few keystrokes that just a few years ago would have taken hours, days, or even weeks for someone who’s worked years and years to grow their art scale to produce.
To an indie author trying to get things rolling, technology like this could transform the industry. If we thought KDP was a shake-up to the publishing industry, this is a whole different ball game.
This is all my opinion, and I still don’t really know what to think.
For businesses, it’s pretty frightening, honestly. But I think creatively it’s kind of amazing. I think ChatGPT and other similar programs (which are bound to be on their way) could be incredible tools for people who have stories to tell.
Anyone can write a story, or draw a picture. But it often takes years of practice to gain and sharpen the necessary skills to bring the image or the story in your mind onto the page. With technology like this, things are a little different. I won’t say the barrier to entry is lower, but it’s definitely different.
I can do some basic photo manipulation, but I’m not an artist. This cover was the first thing I made in DallE2. I watched a few videos on how to use it, but with only a handful of generations, I had images that were basically exactly what I wanted. After that, I spent a couple of hours in Photoshop putting a few of the generated pictures together to make this cover.
What I’m getting at is that soon, I think ChatGPT could be as easy for non-writers to use as DallE2 is for non-artists.
Epilogue: Final Thoughts
While I can’t see this being a huge threat to authors with large fan bases and followings, I think mid-listers and us little guys will see the most change. What exactly those changes are, I don’t know. We’ll just have to wait and find out, I guess.
I also see the possibility of Amazon banning AI content like this in the future, or at least at some point in the future, so if that happens I will put the full ebook here on my website instead.
So if you’d like to read Legends of the Shadow Woods you can check it out on Amazon, and there are some of my other original, non-A.I. generated work there as well.
Stay tuned to my YouTube channel for the latest updates.
-Sydney Faith