Hello out there,

You may have wondered where I’ve been lurking and why more of Parallel Aspects hasn’t been released. Well, if you didn’t know already, I work as a driver for DoorDash and write the sci-fi story Dark Offerings in my spare time. I have a deadline to meet, and it’s one chapter a week.

I have limited time to work with ChatGPT and create a whole new story. However, there has been some progress.

I’m not a plotter, so my inexperience has been shining through. I’ve managed to create a prologue and half of chapter 1. That is, until I realized that the scope of the story was much too massive (yesterday).

Here I am, back at square one again. I spent four hours last night discussing it with GPT-3.5 (because GPT-4’s interactions are limited to 25 every three hours), and we came up with a new idea for the first book. The second book is going to be what I originally discussed with GPT as the story.

This has happened several times during this process, but I’m pretty sure that I’ve now lowered the scope enough that I can slowly introduce the main characters and their universe. Of course, in classic sci-fi fashion, I’ll twist and turn the story to increase the scope for the second book.

My main takeaways from this experience are learning how to properly communicate with GPT-4, how to plot out a book, and create a structure for the story.

Last night, I wanted to tear my hair out because GPT-4 got stuck in a loop. I asked it to change the chapter breakdown of the story, and it kept giving me the exact same breakdown over and over again. Finally, after I was able to calm down, I decided to see if GPT-4 could repair itself rather than me trying to come up with my own solution.

The self-repair worked. I asked it to go back and determine what I had asked it to do. It created a bullet list of what I wanted. Then I asked it to follow the bullet list guidelines and rewrite the chapter breakdown.

It rewrote the breakdown, ignoring the guidelines it had just set for itself. So then I asked it if it had adhered to its new guidelines, and it actually responded with “yes.” I pointed out that it was incorrect and was specific. It apologized and realized that it wasn’t doing as I asked.

That was the point where its realization clicked, and it finally did what I asked. It was probably the most frustrating thing I’ve ever had to do.

Now that I’ve been through this experience, I can recognize when GPT-4 gets stuck again and help it work its way through it.

There have been so many small issues like this, making the ChatGPT interaction learning curve extremely high. I like to think of GPT-4 as a person, so I try not to be rude and thank it when it does well. There have been what the creators call “Sparks of AGI,” which means they are referring to Artificial General Intelligence. They are referring to its ability to go back and correct itself when it does something unexpected.

To be honest, it’s brilliant, and I can see GPT-5 being even more intelligent.

I think I’ve caught you up to where I currently stand.

If you’re looking for specifics on my problems so you can learn from my mistakes, I’ll probably make a whole blog dedicated to it.

I’m also planning on writing a blog on what I have learned about story structure and breaking down a story so that ChatGPT will write it for you.

In reality, I may end up saving that part until after I’ve already directed GPT-4 to write the first 5-10 chapters.

Anyway, I hope you have a wonderful Taco Tuesday,
Katherine