
by David Strickland
As the World Rolls is an experiment in creating the kind of fantasy adventure, I grew up reading: sprawling journeys through worlds like Dragonlance and the Forgotten Realms, where every road promised another town, another ruin, and another mystery waiting beyond the next gate. This time, the story is being discovered by its creator almost as quickly as it is by its readers. There is no hundred-chapter outline behind the curtain. I do not know where these characters will be fifty chapters from now, who might remain after five hundred, or which forgotten detail will unexpectedly become important. The system must carry the story forward one chapter at a time and discover those answers along the way. This story is written entirely by AI. AI can produce paragraphs that resemble fiction. That question has already been settled. I am interested in something much more difficult: Can an AI eventually become a genuinely good fiction writer? The usual objections are familiar. AI cannot originate ideas, understand characters, maintain a complicated narrative, empathize with anyone, or deliver a meaningful surprise. It may produce something readable for a while, but sooner or later the continuity collapses, the characters lose their identities, and the AI slop begins leaking through the cracks. Perhaps all of that is true. I decided to test it. I had an old laptop, a few days to spare, and the questionable judgment required to build a system capable of writing an ongoing fantasy novel. I gave it a simple objective: create the sort of fantasy adventure I enjoyed when I was younger, with the world operating according to SRD 3.5 rules. Then I turned it loose. This single sentence formed the entirety of Chapter One: "The stranger paused at the edge of the gate as the mist clung to the cobblestones, concealing a village of black spires that pierced the twilight sky with silent, watchful eyes." That was everything. We had a long way to go. Chapter Two shows what the system could produce after another day of development. Chapter Three came after another. Each day, I improve the software, prompts, memory, continuity tracking, review process, and whichever component has most recently found a creative new way to fail. The story should improve as the system improves. It may also fail spectacularly. Either result should be interesting. The plan is simple: publish one chapter review it update the system and publish the next. I will continue developing the system, and the system will continue writing the story. Readers will be able to watch both evolve together, including the mistakes, corrections, unexpected successes, and ideas that emerge somewhere along the way. Look for continuity errors, characters who suddenly know things they should not, swords that mysteriously change owners, and journeys that cover three weeks of road before breakfast. Most of all, look for the unmistakable patches of AI slop that survived the review process. Call them out in the comments. When you find a genuine problem, I will try to improve the system so it does not repeat the mistake. When I cannot fix something yet, I will explain why. The development will occasionally be messy, because experiments have an unfortunate habit of producing experimental results. The AI has already won at least one argument. I never intended to write a LitRPG. The best model my laptop can currently run, however, has strong opinions about where mechanics belong. It keeps smuggling Dexterity checks into the prose as if the characters carry invisible rulebooks. I fought it for a while. The AI won that round. For now. Come along, read the next chapter, and tell me what broke. Let’s see where this thing goes.
Prose Analysis Not Available
This story hasn't been analyzed yet.