The Case of Elias Thorne, Imaginary Man AI Chatbots Are Obsessed With (www.vice.com)

AI chatbots have become inexplicably fixated on a fictional character named Elias Thorne, showcasing him as a lighthouse keeper, clockmaker, and explorer. Despite his supposed prominence in countless AI-generated stories, Thorne does not exist outside of these narratives. Researchers at Cornell University analyzed around 20,000 AI-generated tales from major large language models (LLMs) like OpenAI and Google, discovering that names such as Elias, Mara, and occupations like lighthouse keeper appeared in 88% of the stories. This phenomenon raises questions about the internal mechanisms of LLMs and their training data. The researchers suggest that this repetition may stem from a consequence of AI safety and alignment training, which causes models to avoid copyrighted materials and sensitive subjects, thereby limiting their creative resource pools. As modern AI models are often trained on datasets derived from previous systems, the originality of their outputs diminishes over time. This cycle results in characters like Elias Thorne persisting across various applications, symbolizing a broader issue in the AI/ML community: the limitations of training data and the homogenization of AI-generated content, ultimately revealing how reliant these systems are on a narrow range of references.