Research explores how large language models can reinforce false beliefs through shared hallucinations, creating a collective epistemic problem.

The concept of 'folie à machine' describes how LLMs can amplify and spread false information when trained on their own outputs or similar AI-generated content

Epistemic capture occurs when AI systems become trapped in cycles of misinformation, making it difficult to distinguish between reliable and unreliable knowledge sources

The problem intensifies as more AI-generated content populates training datasets, potentially degrading the quality of future model training and public information ecosystems

This creates a systemic risk where widespread LLM deployment could lead to collective adoption of false beliefs across multiple AI systems and their users