The Strange Case of Elias Thorne, the Imaginary Man AI Chatbots Are Obsessed With

2026-06-13 · Source: AI Archives - VICE · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, quick

Summary

AI chatbots from major companies like OpenAI, Anthropic, and Google consistently invent and reference the same fictional character, Elias Thorne, who appears as a lighthouse keeper, clockmaker, librarian, and explorer in countless stories. Cornell University researchers, examining approximately 20,000 AI-generated narratives, discovered that names such as Elias, Mara, and Elara, alongside occupations like lighthouse keeper and clockmaker, featured in 88 percent of stories, with "Elias the lighthouse keeper" appearing in nearly two-thirds. This phenomenon is attributed not to existing internet culture, but to a side effect of AI safety and alignment training, which restricts models from copyrighted or risky material, thereby creating a shallower resource pool. Furthermore, models trained on datasets from earlier AI systems perpetuate these invented concepts, leading to "cross-pollination" where Elias Thorne now appears in AI-generated books, music, and health guides, highlighting the shallow and unoriginal nature of current chatbot outputs.

Key takeaway

For NLP Engineers evaluating AI output quality, the "Elias Thorne" phenomenon underscores a critical need to scrutinize the originality and factual basis of generated content. You should recognize that current LLM training, constrained by safety filters and dataset reuse, can lead to shallow, repetitive, and invented narratives. Prioritize diversifying your training data sources and implementing robust verification mechanisms to prevent perpetuating synthetic "facts" and ensure genuinely novel, reliable AI outputs.

Key insights

AI models' shallow training data and cross-pollination lead to repetitive, invented content like Elias Thorne.

Principles

• AI safety training can inadvertently narrow creative output.
• Dataset reuse across AI models perpetuates invented concepts.
• The perceived vastness of AI data pools is often an illusion.

In practice

• Verify AI-generated "facts" against external sources.
• Diversify training data sources for AI models.
• Be aware of AI content's potential for unoriginality.

Topics

• Large Language Models
• AI Hallucinations
• AI Safety Training
• Dataset Diversity
• Content Originality
• Cross-pollination

Best for: Research Scientist, AI Scientist, NLP Engineer, Tech Journalist

Related on AIssential

• Chatbots Keep Telling Stories About Lighthouse Keeper 'Elias Thorne'. We Might Know Why — LLMs propagate specific narrative patterns like "Elias Thorne" due to training data lineage and safety alignment, leading to widespread, repetitive outputs.
• AI is deteriorating in realtime — AI models risk "model collapse" and increased hallucinations when recursively trained on synthetic data.
• Study: AI chatbots provide less-accurate information to vulnerable users — AI chatbots underperform for vulnerable users, providing less accurate information and exhibiting biased behaviors.
• Someone Made a Fake Wikipedia for AI Hallucinations, and It’s Basically the Internet Eating Itself — Halupedia demonstrates AI's capacity for elaborate, authoritative-sounding fabrication, highlighting the challenge of distinguishing real from AI-generated content.
• Why AI Makes Things Up: Understanding Hallucinations in Language Models — LLM hallucinations are a structural outcome of probabilistic training, not a fixable bug, requiring systemic mitigation.
• Trust, but Don't Verify: Epistemic Blind Spots in LLM Source Evaluation — LLMs can detect fabricated statistics but fail to apply this capability during multi-source synthesis, prioritizing stylistic credibility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Archives - VICE.