Why ChatGPT, Gemini, & Claude Keep Writing About The Same Fictional Character Named Elias Thorne

Researchers found that AI chatbots repeatedly generate the same fictional names, including Elias Thorne. The pattern may be linked to AI safety training and recycled training data.

Same invented man keeps appearing across platforms, raising real questions about how original AI truly is.

Source : ABPLIVE AI

Show Quick Read

Key points generated by AI, verified by newsroom

• Fictional Elias Thorne often appears in diverse AI-generated content.
• Study revealed few specific names dominate AI-generated content.
• AI safety training and data reuse cause content repetition.

Ask any major AI chatbot to write a short story, and there's a good chance a man named Elias Thorne will show up. He's been cast as a lighthouse keeper, a clockmaker, a librarian, and an explorer across countless AI-generated tales, books, YouTube videos, and even health guides. 

On paper, he sounds like one of the most written-about men alive. Except he isn't real, and nobody is entirely sure why every chatbot seems to know him.

What Did Researchers Find About AI's Repeated Names And Characters?

According to a report by 404 Media, researchers at Cornell University set out to understand this pattern. The team studied around 20,000 AI-generated stories produced by major language models from OpenAI, Anthropic, and Google, and found a strikingly small set of names and professions repeating across the board. 

Words like Elias, Mara, Elara, lighthouse keeper, clockmaker, and librarian turned up in 88% of the stories analysed, with Elias the lighthouse keeper alone appearing in nearly two-thirds of them.

The first assumption was that these names came from a specific book or some corner of internet culture that the models had absorbed during training. However, the researchers found no real evidence to back that up. Their theory instead pointed to AI safety and alignment training as the likely cause.

Could AI Safety Training Be Behind This Pattern?

AI companies train their models to avoid generating copyrighted material, partly to stay clear of legal trouble with large entertainment companies, and to filter out risky or adult content. This kind of training narrows the pool of material a model can safely draw on when generating new text. Add to this the fact that newer AI systems are frequently trained on data generated by older AI systems, and the cycle becomes self-reinforcing, with little room for fresh variation.

This may explain why, once Elias Thorne appeared somewhere in the training data, he kept resurfacing across different models and platforms. Tracing his exact origin might be impossible at this point. Software engineer Daniel May even spotted the name appearing in dubious health guides, alongside its presence in AI-generated books, music listings on Amazon, and YouTube videos.

What looks like an endless well of information that AI models pull from is, in practice, fairly shallow. Once a system has consumed most of the readily available data, repetition sets in. Elias Thorne may just be a strange glitch, but he also says a lot about how repetitive AI-generated content can become without genuinely new input.

Why Meta Employees Are Feeling Less Motivated As Company Doubles Down On AI