A recent study conducted by Cornell University analyzed 20,000 fictional stories generated by four different language models in May 2026. The research revealed a striking pattern of narrative repetition, focusing particularly on the character named Elias. The name elias appeared in approximately a quarter of all analyzed stories, a frequency noted to be about 900 times higher than its occurrence in actual published literature.
The analysis highlighted that certain combinations of words were disproportionately common across the generated texts. Specifically, the same eleven words—including the names Elias, Mara, and Elara, alongside professions such as lighthouse keeper, jeweler, baker, and librarian, and the location of a lighthouse—were present in 88% of the stories. Furthermore, the iteration of Elias as a lighthouse keeper appeared in nearly two-thirds of the generated content, suggesting a highly popular and predictable narrative archetype within the AI outputs.
Researchers theorize that this uniformity stems directly from the data used to train the artificial intelligence. The findings point toward a dataset known as WildChat, which comprises one million real conversations with ChatGPT. This suggests that the models are heavily influenced by the specific patterns and tropes embedded within their learning material, leading to the most consistent narrative outcomes.
The study provides insight into how the training data dictates the most frequently generated elements in artificial storytelling.
Topics: #most #elias #popular
