This repository serves as a record of an interesting AI experiment in which I generated a one hundred plus page travel memoir (fictitious!) based on a single prompt to Sonnet 3.7,
| File | Description |
|---|---|
| Generated Text | The AI-generated story output from the experiment (unedited) |
| System Prompt | The system instructions given to the AI model |
| User Prompt | The creative prompt containing the story premise |
| Prompt Script | Python script used to format and send prompts to the AI |
| Experiment Data | Metrics and observations from the experiment |
| Ebook Cover | Cover image for the generated story |
Around April 10, 2025, I found myself delving into the exciting advancements of Google's Gemini 2.5, which had been launched about a week prior. I was particularly struck by the considerable increase in the maximum continuous output capabilities of the model. This got me thinking: could an AI conceivably write an entire book from a single prompt?
To put this to the test, I crafted a system prompt along with a basic plot loosely inspired by my own emigration from Ireland to Israel – the only difference being the notable absence of a magical, talking sloth in my real-life experience! The results I obtained were quite unexpected.
While Gemini didn’t quite reach its advertised maximum output, let alone come anywhere close to it (details in the data doc), I discovered that Sonnet 3.7 (most famous, currently, for its agentic code generation abilities) promises a staggering 128,000 tokens with specific API call formulation. Although I didn’t manage to coax anything close to that figure, or even the 40,000 words I requested in my prompt, the output I received significantly surpassed what Gemini had provided.
The generated work of fiction is quite a fascinating outgrowth of my prompt, I think. With minor deviations here and there, the characters as prompted are held mostly consistently throughout the text leading up to its natural conclusion, and the AI demonstrated an impressive ability to create independently subplots and narratives that were not explicitly stated in the body of the prompt itself.
While its word count means that it falls in a literary gray zone, it's almost .... sort of a short book.
The text, as well as some AI authors appendages to it are included here as an archive of this experiment.