AI Chatbots Repeat 'Elias Thorne' Story Due to Training Data

A recurring character named Elias Thorne, along with similar themes like lighthouse keepers and clockmakers, is appearing in a significant percentage of stories generated by major AI chatbots. Researchers suggest this phenomenon stems from the models' safety tuning and the interconnectedness of their training data, with OpenAI's GPT-3.5 acting as a foundational source for many subsequent models. This linguistic pattern has spread beyond chatbots, with Elias Thorne appearing as an author in AI-generated books on Amazon across various genres, including potentially harmful topics, and in AI-generated YouTube content. AI

IMPACT Reveals a potential mode collapse in LLMs, impacting the perceived creativity and uniqueness of AI-generated content.

RANK_REASON The cluster reports on a research paper analyzing a phenomenon in AI chatbot outputs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on 404 Media →

• Anthropic
• ChatGPT
• Cornell University
• David Mimno
• Deepseek
• Elias Thorne
• Gemini
• Google
• GPT-3.5
• Grok
• OpenAI
• Rebecca M. M. Hicke
• Sil Hamilton
• WildChat

• model release
• safety
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →