The curious case of Elias Thorne – and what he tells us about AI inbreeding (www.theguardian.com)

In a striking observation, researchers from Cornell University have uncovered a peculiar trend involving a character named Elias Thorne that frequently appears in stories generated by large language models (LLMs) like ChatGPT and Claude. After analyzing 20,000 AI-generated narratives, they found that 26.5% included Elias, often depicted as a lighthouse keeper or clockmaker. This prevalence raises questions about the underlying mechanics of AI storytelling and hints at a broader phenomenon of "model collapse," where reliance on limited datasets leads to repetitive outputs. The significant takeaway for the AI/ML community is the implication of AI inbreeding, where models trained on similar stories perpetuate the same characters and contexts, thus diminishing creative diversity. Researchers speculate this fixation might stem from AI's training restrictions to avoid copyrighted content, resulting in a narrow pool of inspiration. Furthermore, the emergence of Elias Thorne beyond AI-generated content—such as in self-published books and YouTube videos—highlights the potential for low-quality outputs to proliferate in the digital landscape, which could ultimately affect the future development and quality of AI models. This trend serves as a cautionary tale about the interconnectedness of AI outputs and the importance of diverse training data.