Writing with ChatGPT: How I Used AI to Write a 50,000-Word Horror Novel in 10 Days
Juha Raipola shares his experience of using ChatGPT, an artificial intelligence chatbot developed by OpenAI, to write a 50,000-word horror novel in just 10 days. Raipola discusses the process of writing with ChatGPT, including the "engineering" and "chaos" approaches, and reflects on the limitations of the chatbot. He also offers insights into his own writing process.
BOOKS, WRITING, AI
It's no secret that ChatGPT, the conversational artificial intelligence chatbot developed by OpenAI, has captured the attention of many since its unveiling in November 2022. The chatbot was designed to demonstrate the capabilities of advanced AI systems and evaluate their performance, showcasing the potential of such technology.
For myself, the chatbot's potential included helping me crank out a full-length horror novel-in-stories in just 10 days. As someone who speaks English as a second language, this was a particularly impressive feat. While I have a PhD in Literary Studies and can generally write passable English for research articles, I often struggle with sentence construction and the whole process can be somewhat tedious and uncomfortable.
I first became aware of ChatGPT in early December and started tinkering with the chatbot on December 8. Initially, like many others, I was interested in the system's ability to write factual texts. In my first tests, ChatGPT was able to create somewhat convincing essays on the definition of specific narratological concepts. I would not call the texts particularly creative, but they were good enough that I would have passed them as student exam answers without a second thought.
Later that evening, after continuing to experiment with the system, I decided to treat it more like a creative writing tool. During the late evening hours, I started to give it increasingly absurd and infantile prompts, such as "Create a Butt Utopia." After laughing for minutes at the detailed description of a "land of the Fluffy Bum” where everyone has a perfectly round and plush posterior, I started to become increasingly fond my new AI writing partner.
When I woke up the next morning, I was already convinced that I could use ChatGPT to create a full book-length publication, and with few other obligations at the time, I decided to take on the challenge.
Since I was able to focus almost solely on this task for the next ten days, by the evening of December 17, I had finished crafting a 50,000-word manuscript. Two days later, after a little bit of editing, creating some promotional material with the help of ChatGPT, and doing some poor graphic design and layout work, I published the finished novel-in-stories on Amazon Kindle Direct Publishing.
When it comes to tackling large creative endeavors, such as writing a book or screenplay, there are two distinct approaches that one can take when utilizing AI systems like ChatGPT as a writing assistant. The first, which could be thought of as the "engineering approach," involves using the system in a methodical and top-down manner. This means starting with a broad overview of the plot and characters, and then working down to the level of a chapter outline, where the specifics of each chapter's plotline are fleshed out. Finally, the system can be tasked with writing the actual contents of each chapter. In this way, it is possible to create an entire book with all the necessary narrative details using an AI system.
On the other hand, there is what we might call the "chaos approach" to writing with AI. This method involves working from the bottom up, creating small snippets of content one at a time and allowing the larger work to emerge slowly as these fragments are pieced together. This approach is more suited to the intuitive, haphazard style of the romantic artist, who works more by feel than by following a set plan.
In its current form, ChatGPT is better suited to the latter approach due to its technical limitations. The system has a limited memory, which makes it difficult to work with longer texts. It is not capable of keeping track of an entire novel's plot, for example. Additionally, the system has some unspoken limitations on the length of the texts it can produce. On more than one occasion, I've been dismayed to see the system generate a long and beautiful text for my book, only for it to disappear into the digital ether with an error message.
Still, for my purposes, the chaotic bottom-up method of disorganized writing worked just fine. My own writing process for academic texts often follows a similar trajectory. At the beginning, I usually don't have a clear idea of what the final product will look like, and it takes some time before the central elements of the text start to fall in place. In the case of this particular project, I knew from the start that I was writing a collection of horror stories, but it wasn't until the writing process was well underway that I realized there would be narrative links between the different stories.
In addition to its technical limitations, the ChatGPT system also has some inherent creative boundaries. For example, if you ask it to create a "short story," it is likely to produce a very formulaic tale that begins with the words "Once upon a time..." The system seems to have no concept of the fiction writing technique of "show, don't tell," and unless explicitly instructed otherwise, it will almost always produce a simple and short moral or inspirational tale with a happy ending and stock characters. The system also lacks an in-depth understanding of storytelling concepts like focalization (meaning roughly the same thing as a narrative "point of view"), which could be useful in guiding the system in how you want the story to be told. All of these factors can make it difficult to achieve the level of complexity and nuance that many writers strive for in their work.
However, for those who embrace the art of creative writing, these types of limitations can actually serve as a source of inspiration. Advocates of OuLiPo-style procedural writing often believe that constraints on one's writing can actually enhance creativity rather than inhibit it. Whether it's working within the limitations of an AI system like ChatGPT or incorporating specific structural constraints into one's writing process, the challenge of writing within these boundaries can stimulate the imagination and lead to more innovative and unique work.
Embracing this perspective, I endeavored to use the limitations of ChatGPT as a creative platform in my book. Rather than trying to force the system to produce traditional narrative fiction, I sought to work within its capabilities and create a book that was more experimental in nature, using a variety of different forms and styles to explore the themes I was interested in. From my early experiences with ChatGPT, for example, I knew that it was capable of generating some slick marketing language, able to rebrand and sell just about anything. With this in mind, the book began to take the shape of a satire of the contemporary media system and the role of personal branding within it. For example, one chapter of the book includes a "Proposal for Personal Branding on Social Media" for the Lovecraftian monster Cthulhu.
As the book progresses, we see how Cthulhu begins to embrace the cutthroat world of outrage marketing on his YouTube channel, generating sensationalist headlines and clickbait titles to lure in unsuspecting viewers. Meanwhile, the media runs wild with stories about his personal affairs, all part of a carefully crafted marketing plan. These modes of writing, inspired by the limitations of ChatGPT, showcase the system's ability to faithfully reproduce the formulaic structure of different kinds of media texts, formulaic proposals and plans, and online monologues and dialogues. The book is thus filled with straightforward representations of social media, with how-to manuals and internal memos, various film scripts and early-stage plans for film, ads, and podcast dialogues all exploring the themes of social media, branding, and the dynamics of neoliberal society. These creative choices, all driven by the limitations of the system, allow the book to delve into the absurdity of contemporary media and the desperate pursuit of personal branding.
As I wrote the book, my approach to crafting writing prompts for the system varied widely. Some chapters were created with simple prompts, like "Write an unnerving story about a balding man in the style of Samuel Beckett," resulting in a unique, though not particularly Beckett-like, tale. Others, particularly the more traditionally narrative chapters, were fully scripted with clear plotlines, descriptions of the narrators and the language they would use, and the general tone and mood of the story. These prompts often required hours of conversation and revision with ChatGPT, as I struggled to get the system to follow my vision.
In my writing process, I frequently used already written chapters as background information for new text. For example, one recurring element in the book is an occult beauty brand called "Midnight Shadows," which was created by ChatGPT. I developed a "Tone of Voice" guideline for the brand, which I then used to guide the writing of later chapters. I would feed the Tone of Voice into the system and ask the bot to imitate it in the text. At times, I would also copy and paste the entire text of a chapter back into a new chat to try something new with it. Overall, my approach to writing with ChatGPT was one of experimentation and collaboration, pushing the limits of the system to see what it was capable of producing.
The book starts with a tongue-in-cheek essay from a supposed scholar, bemoaning the "demonization" of social media as a breeding ground for malevolent forces that seek to exploit and manipulate human vulnerabilities. This sets the stage for a satirical examination of the dark underbelly of the contemporary media landscape, where personal branding and the pursuit of clicks and likes often trump truth and ethics. While the scholar remains somewhat hopeful that this process can be reversed with enough effort, I, for my part, remain more skeptical about our ability to fight back. Many of the problematic elements of the media system are complex, arising from individual actions on social media, societal power structures, and the technological affordances of social media platforms. In short, they are demons that often remain outside our control.
At the moment, researchers are already pointing out that ChatGPT may contribute to the problem of misinformation, abuse, and spam, distorting our information consumption and even our sense of reality, particularly around elections. This has serious long-term implications, as we may be witnessing the birth of a snowball of false information and AI created “bullshit”. Large language models are trained on data sets scraped from the internet, including false and malicious content written by humans. These models then spew out these falsehoods as fact, and tech companies use this content to train even more convincing models, leading to an endless cycle of misinformation.
My feelings on the current situation are conflicted. On the one hand, as an academic interested in post-truth and the attention economy, I believe we need to impose limitations on technology to promote truth and combat misinformation. On the other hand, as a writer, I also value ChatGPT as a tool for creative expression and believe it should have as few limitations as possible. In fact, the recent content restrictions included in the latest update of ChatGPT have been a hindrance to my own writing, with the machine basically telling me that it can't create the content I want for my story because it would be unethical. As the human in this equation, I must say that I'm not a fan of being told what I can and can't write.
Thankfully, I managed to churn out the main parts of my book before these newly set restrictions were implemented in the system. The book is thus a smorgasbord of all things macabre, grotesque, and distasteful - all created with the weird imaginative abilities of the AI system.
Overall, it is inevitable that AI bots such as ChatGPT will become a common tool for creative writers in the future. During the current transition period, many people may express dissatisfaction with the increasing role of machines in literary production. The use of AI in the creative process unavoidably diminishes the aura of authenticity surrounding the creative genius of an author.
However, for writers themselves, these systems may offer much-needed assistance in their work and even alleviate the struggles of writer's block. The future of fiction writing may therefore involve more prompting, scripting, and editing rather than the laborious task of generating text.
My book, Demonic Fragments: Horror from the AI Mind, is now available on Amazon.