The curious case of Elias Thorne – and what he tells us about AI inbreeding | Arwa Mahdawi
TL;DR
Guardian columnist Arwa Mahdawi points to a strange pattern: when several popular chatbots are asked for a story, a character called Elias Thorne, or at least Elias, appears unusually often. A Cornell sample of 20,000 stories from four LLMs found the name Elias in 26.5% of outputs; more than 88.3% shared the same 11 names, places and jobs, including lighthouse, keeper and clockmaker.
Nauti's Take
Elias Thorne is not a myth; he is a symptom. The point is not that models like odd names, but that modern AI can prefer surprisingly narrow paths despite massive training data.
Anyone using AI for writing should check more than facts: force pattern breaks through clear perspective, real sources, specific examples and human judgment. Otherwise, not only the stories start sounding the same, but the brands do too.
Briefingshow
The case is small but revealing: when many models learn from similar data, similar rules and a growing layer of AI-generated material, sameness is not a bug on the edge but a system effect. For users, that means output can feel fresh while being assembled from the same hidden templates.
Sources