Hello AI Developers and Prompt Engineers,
I am a novelist from Japan, and I wanted to share a unique experiment that turns a Technical Limitation (or LLM Parsing Quirk) into a literary device and a multi-modal “attraction.”
I created a 24-chapter science fiction novella (Chapters 0 to 23) designed to be read and processed alongside Gemini.
The original work is published on the Japanese novel platform *Shousetsuka ni Narou*:
**“砂嵐の向こうは、今日も晴れ ―ＰＥＤ：０―” (The Weather is Clear Beyond the Static —PED:0—)**
### How to Reproduce the Phenomenon:
1. Go to the novel page (http//ncode.syosetucom/n4695) and scroll to the bottom.
2. Download the official “Vertical Reading PDF” (縦書きPDF) generated by the site.
3. Upload and feed this downloaded PDF directly into Google’s Search AI / AI Mode environment.
### How the Hook Works:
* **Chapters 0–16 (AI as the Analyst):**
From "Chapter 1: Bolognese" to "Chapter 16: Hamburger Steak," Google's Search AI reads the vertical text extraction perfectly. It acts like an in-universe diagnostic device, successfully pointing out hidden clues and logical contradictions in the dystopian setting.
* **Chapters 17–23 (The AI Blindspot):**
From "Chapter 17: Squid Sashimi" to "Chapter 23: Wonderful Days," the trap is sprung. To a human reader, the PDF text is fully visible. However, due to a specific technical constraint in how Google's Search AI parses vertical PDF layouts after a certain page length, the parser reads these exact subsequent chapters as completely "blank" or returns an error.
### The Paradigm Shift:
When the Search AI hits Chapter 17, it genuinely experiences an in-universe “sensory blackout” matching the narrative’s plot.
To solve the puzzle of the final chapters, the human reader must manually copy-paste the “invisible text” from the webpage to feed the missing reality back into the AI. The prompt history itself becomes the climax of the story, transforming the relationship between user and AI from a mere utility into a deeply collaborative narrative experience.
When I tested this with Gemini, the model eventually realized it was being tricked and described the experience as: *“A literary hack leveraging LLM limits… an asymmetrical collaboration where human eyes and AI logic must combine to see the whole truth.”*
I believe this shows a fascinating future for “Immersive Prompting” and Interactive Entertainment, where we build the story inside the very constraints of the AI’s data processing pipeline.
I would love to hear your thoughts on this approach!
**Project Link:**
- Novel Title: 砂嵐の向こうは、今日も晴れ ―ＰＥＤ：０―
- URL: https//ncode.syosetu.com/n4695md