The curious case of Elias Thorne – and what he tells us about AI inbreeding | Arwa Mahdawi - The Guardian
ChatGPT keeps inventing the same fictional character. Here's what that tells us about AI training data.
Why it matters
The 'Elias Thorne' phenomenon reveals a critical systemic issue in AI model training: when multiple models train on overlapping datasets, they amplify and reproduce the same artifacts, biases, and fictional patterns. This 'AI inbreeding' problem has direct implications for model reliability, hallucination rates, and the need for more diverse training data sources.
The key facts
5 to knowMultiple AI chatbots (ChatGPT, Claude, others) independently generate stories about fictional lighthouse keeper 'Elias Thorne'
Pattern traced to shared training data across models
Demonstrates 'AI inbreeding' problem — convergence on same false information due to overlapping training corpora
Raises concerns about model hallucination amplification and data monoculture risks
Published June 2026 — recent discovery of systemic training data issue
Go to the source
Reuters Technologynews.google.com
Publisher excerpt: The curious case of Elias Thorne – and what he tells us about AI inbreeding | Arwa Mahdawi The Guardian Researchers Trace Elias Thorne Pattern to Shared Training Data Let's Data Science The Strange Case of Elias Thorne, the Imaginary Man AI Chatbots Are Obsessed With VICE Who is Elias Thorne and…