A curious phenomenon has emerged in the world of artificial intelligence: a character named Elias Thorne appears in a remarkable number of chatbot-generated stories. This figure, whether portrayed as a lighthouse keeper, clockmaker, or detective, has become a staple in tales produced by popular large language models (LLMs) like ChatGPT and Claude.
The Ubiquity of Elias Thorne
Recent research from Cornell University sampled 20,000 stories from four LLMs, finding that the name Elias appeared in 26.5% of them. Furthermore, over 88.3% of generated stories shared the same 11 names, locations, and professions, including lighthouse, keeper, and clockmaker. This suggests a narrow pool of inspiration for AI storytelling.
Why Is AI Obsessed with Elias?
While the exact reason remains uncertain, the Cornell paper speculates that AI models may have been instructed to avoid copyrighted characters and adult content, leading them to rely on a limited set of references. Additionally, AI models learn from each other, causing quirks like the Elias fixation to spread rapidly. One researcher described it as a virus.
Beyond Fiction: Elias Thorne in the Real World
The character has moved beyond AI fiction, appearing as a byline on dubious self-published books on Amazon and in AI-generated YouTube videos. This spread may indicate a phenomenon called model collapse, or AI inbreeding, where AI learns from its own low-quality output, leading to increasingly poor results.
The Danger of Model Collapse
As more of the internet becomes filled with AI-generated content, future models will train on this slop, perpetuating a cycle of degradation. This process mirrors capitalism, containing the seeds of its own destruction. However, experts warn that AI may cause significant harm before collapsing.
Elias Thorne serves as a warning about the limitations and risks of generative AI, highlighting the need for careful management of training data and model interactions.
