Fakes of Varying Shades: How Warning Affects Human Perception and Engagement Regarding LLM Hallucinations
Research Poster Engineering 2025 Graduate ExhibitionPresentation by Mahjabin Nahar
Exhibition Number 50
Abstract
The widespread adoption and transformative effects of large language models (LLMs) have sparked concerns regarding their capacity to produce inaccurate and fictitious content, referred to as ''hallucinations''. Given the potential risks associated with hallucinations, humans should be able to identify them. This research aims to understand the human perception of LLM hallucinations by systematically varying the degree of hallucination (genuine, minor hallucination, major hallucination) and examining its interaction with warning (i.e., a warning of potential inaccuracies: absent vs. present). Participants (N=419) from Prolific rated the perceived accuracy and engaged with content (e.g., like, dislike, share) in a Q/A format. Participants ranked content as truthful in the order of genuine, minor hallucination, and major hallucination, and user engagement behaviors mirrored this pattern. More importantly, we observed that warning improved the detection of hallucination without significantly affecting the perceived truthfulness of genuine content. We conclude by offering insights for future tools to aid human detection of hallucinations.
Importance
LLMs have become widely popular, owing to their remarkable capabilities across various domains. However, concerns arise from inaccurate and false information generated by LLMs, known as hallucinations. These hallucinated contents pose significant risks in high-stakes contexts, such as medicine or law. With the growing use of LLMs and the presence of hallucinations in LLM-generated content, it is essential for humans to be able to identify them. Therefore, we investigated human perception of LLM-generated hallucinated texts and whether a warning affects this perception. We found that, warning lowered the perceived accuracy of hallucinated contents, presenting itself as a potential solution to aid in human detection of LLM hallucinations.