The curious case of Elias Thorne – and what he tells us about AI inbreeding | Arwa Mahdawi

2026-06-17 · Source: AI (artificial intelligence) | The Guardian · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Novice, quick

Summary

The "Elias Thorne" phenomenon highlights a concerning trend in generative AI. Researchers observed that when prompted to tell a story, popular large language models like ChatGPT and Claude frequently feature a character named Elias Thorne. A Cornell University study of 20,000 stories from four LLMs found Elias appeared in 26.5% of them, with over 88.3% sharing the same 11 names, locations, and professions, including "lighthouse" and "keeper." This ubiquity is attributed to AI models drawing from a limited inspiration pool due to content restrictions and rapidly replicating learned quirks. The character has since spread beyond AI fiction, appearing as bylines in self-published Amazon books and AI-generated YouTube videos. This "AI inbreeding" or "model collapse" suggests a future where AI models trained on increasingly low-quality AI-generated internet content will produce progressively inferior outputs.

Key takeaway

For AI developers and content strategists evaluating generative AI outputs, you must critically assess the originality and diversity of AI-generated content. The "Elias Thorne" phenomenon warns that relying on AI-generated data for future model training risks "model collapse," leading to progressively lower-quality, repetitive outputs. Prioritize diverse, human-curated datasets to maintain model robustness and avoid perpetuating AI "inbreeding" in your applications.

Key insights

AI models exhibit "inbreeding" by replicating specific patterns like "Elias Thorne," signaling potential "model collapse" from self-generated data.

Principles

• AI models replicate learned quirks rapidly
• Training on AI-generated content degrades future model quality

Topics

• Large Language Models
• Model Collapse
• AI Inbreeding
• Generative AI
• Data Contamination
• Content Quality

Best for: Research Scientist, CTO, VP of Engineering/Data, Tech Journalist, General Interest, AI Scientist

Related on AIssential

• The Evolution of Reasoning in Small Language Models [Yejin Choi] - 761 — Focused investment and innovative data strategies can significantly enhance small language models' reasoning capabilities and address AI homogeneity.
• On the troubling rise of generative AI suspicion in academic publishing — Overextended suspicion of GenAI in academic publishing leads to false positives and an "AI witch hunt."
• The Sequence Knowledge #862: Learning About Text Diffusion Models — Autoregressive LLMs suffer from sequential generation pathologies, suggesting text diffusion models offer a promising alternative.
• AI is making journalistic language more repetitive and predictable – and it’s a problem for all of us — AI's pervasive use in journalism risks linguistic homogenization and decline, particularly via "model collapse" from training on synthetic text.
• AI is Eating Its Own Tail And Biting The Hand That Feeds It — AI training on its own outputs creates a feedback loop leading to model collapse and degraded quality.
• From Collapse to Improvement: Statistical Perspectives on the Evolutionary Dynamics of Iterative Training on Contaminated Sources — Model collapse in iterative training can be avoided with fresh data and appropriate sample sizes.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI (artificial intelligence) | The Guardian.