How I Prompted This Hard Sci-Fi Story with ChatGPT 4.5

One night I was experimenting with ChatGPT to see if I could prompt a hard sci-fi short story that was constrained in a similar manner to Alastair Reynold’s ‘Revelation Space’ Universe. That means no FTL (Faster Than Light) travel, decades-long journeys, and scientific details that would hold up under scrutiny. I used GPT-4.5 for this task and started the process with ChatGPT’s deep research feature on nearby stars, exoplanets, and other scientific details for the story.

My Approach

• Constraints First: I told the model to keep it hard science (real exoplanets, real stars, and relativistic flight only), include a shipboard AI that plays a key role throughout the story, and structure the narrative so it includes three mini-chapters (Earth Departure → Proxima Centauri → Next Best Star System).
• Deep Research: I utilized deep research to gather and evaluate scientific facts for the story. Exoplanets featured include:
  • Proxima b (~1.07 M⊕, ~0.048–0.05 AU, ~11.2-day orbit, likely tidally locked; harsh flare environment).
  • Ross 128 b (~1.3–1.4 M⊕, ~0.049 AU, ~9.9-day orbit) orbiting a quieter red dwarf—better odds for a stable atmosphere and retained biosignatures.
• Realistic First Contact: I found the first draft of Chapter 3 very unrealistic as the AI was able to instantly act as a translator for the extraterrestrials, so I prompted ChatGPT to revise Chapter 3 so the crew and AI had to gather a real corpus of signals and behaviours before meaning emerged. On the cloud-shrouded world (permanent overcast), “never seeing the stars” became both biologically and culturally coherent.

​Why Use An LLM For Hard Sci-Fi?

​GPT-4.5 handled style and cohesion while I provided the outline, anchored the physics, and directed the story. The model synthesized tone, pacing, and scene transitions, while deep research grounded the numbers, orbits, stellar behaviour, and environments. That split kept the story tight and credible.

Pattern Recognition, Language, And New Environments

The translation arc I wanted in the story mirrors how real understanding grows:

• Data Before Meaning: You don’t decode a language from one utterance; you need hours of audio, context, and interaction.
• Multimodal Grounding: Sound plus behaviour plus setting beats raw text or speech alone.
• Probabilities, Not Certainties: Hypotheses get proposed, tested, and revised—exactly how good science (and careful field linguistics) works.

AI Is Already Accelerating Science

This workflow reflects where AI is useful today:

• Astronomy: Classifying light curves, denoising spectra, flagging exoplanet candidates, modeling stellar flares.
• Planetary Science: Terrain mapping, autonomous navigation, onboard triage for what to study next.
• Signal Analysis/SETI-Adjacent: Anomaly detection and structured-signal characterization.

​Read The Story

​If you’d like to see the finished product of this experiment, you can read the full three-chapter short story here:

Interstellar Odyssey (PDF)
File Size:	219 kb
File Type:	pdf

Download File

---

Your browser does not support viewing this document. Click here to download the document.

The story follows humanity’s first interstellar voyage—beginning with Earth’s departure, continuing through the Proxima Centauri system, and culminating in first contact on Ross 128 b.

​​Image generated with ChatGPT. 

​Takeaways

If you want to prompt great stories with ChatGPT:

• Start with clear constraints.
• Utilize verified data. 
• Let the model handle cohesion and voice. 
• Keep a human in the loop for truth and taste. 

The result is fiction that reads smoothly but leans on real numbers—a voyage carried by patience, pattern recognition, and the quiet partnership between humans and their machines.