88% of AI-Generated Stories Rely on Just 11 Words

Uncovering the Creative Limits of Large Language Models

According to НВ — Техно: On June 12 at 12:00, researchers revealed that leading AI language models-including GPT-5.4 Mini, Claude Haiku 4.5, and Gemini 3.1 Flash-Lite-produce stories using a strikingly narrow vocabulary in 88% of cases. This core set contains only 11 words:

• Lighthouse
• Keeper
• Baker
• Mayor
• Clockmaker
• Fisherman
• Librarian
• Conductor
• Mara
• Elias
• Elara

The character Elias-often portrayed as a lighthouse keeper-appears in roughly two-thirds of all generated narratives, making him the most frequent figure across these stories.

Software engineer Daniel May first flagged this pattern. Together with researchers Sil Hamilton and David Mimno, he analyzed nearly 20,000 stories produced by the aforementioned models. Despite Elias’s prevalence, the team found no evidence that this character appears unusually often in literature or training data. They hypothesize that his popularity may stem from the WildChat dataset, which includes millions of human-chatbot dialogues based on GPT-3.5.

Key Details About the Character Elias Thorne

Further investigation by 404 Media uncovered that the name Elias Thorne appears in fantasy books and also belongs to an ambient music artist on Amazon. Daniel May additionally discovered that among books listing Elias Thorne as the author, one is a guide to alternative cancer treatments. These findings suggest that the AI’s favored character may be rooted in specific data present in the models’ training sets.

This research into AI language models highlights the critical need to examine the data these systems learn from. Such lexical constraints and recurring character tropes may point to systematic biases in training datasets, ultimately shaping the creativity and diversity of machine-generated stories.

As researchers delve deeper into the limitations of AI-generated narratives, it's essential to consider other instances where artificial intelligence has stumbled. For example, a recent analysis revealed that Google's AI struggled with basic tasks, raising questions about the reliability and accuracy of these systems. Understanding these challenges can provide valuable insights into the broader implications of AI in creative fields.