GPT-5.5, Claude, and Gemini Share Stable Fiction Preferences Including 'Resonances and Echoes'
Researcher Ethan Mollick documented that GPT-5.5 exhibits stable, identifiable fiction preferences: lighthouses, the ocean, maps, bells, clock towers that ring impossible times, the name "Mira Vale," resonances and echoes, and "secret third things." These preferences have been stable across versions—not a new GPT-5.5 quirk. Notably, Claude and Gemini independently share the "resonances and echoes" pattern, suggesting cross-model convergence on certain literary aesthetic tropes despite different training pipelines.
Why It Matters
Stable, identifiable creative fingerprints across frontier models—including cross-model convergence—suggest training data and RLHF processes are producing systematic aesthetic biases that persist across versions and labs, with implications for AI-generated content detection and editorial authenticity.