Chatbots give a narrow slice of knowledge:Researchers warn of knowledge collapse (news.ku.dk)

A recent study led by researchers at the University of Copenhagen has revealed that AI chatbots provide a significantly narrower range of information compared to traditional web searches. Analyzing 27 large language models across 155 topics, the study found that even the most diverse model tested, OpenAI’s GPT-5, generated at least 18.7% less variety in responses than Google searches. This uniformity raises concerns about a potential "knowledge collapse" where users become increasingly exposed to repetitive and limited perspectives, ultimately diminishing the richness of information available to them. The study underscores the implications of relying on AI models for knowledge acquisition, particularly as the models are designed to prioritize patterns from the most frequently occurring content. The researchers caution that if future language models are predominantly trained on outputs generated by their predecessors—already less diverse than human-created texts—this could perpetuate a cycle of reduced diversity over time. They encourage both users and AI developers to remain vigilant and seek varied sources of information to ensure broad perspectives and prevent a decline in the richness of accessible knowledge.