Data from 199 residents of the United States comparing reported trust in output generated by a large language model and output provided via a different form of artificial intelligence (January 2024)

This data was collected in an experiment aiming to establish whether trust in large language models (LLMs) may be inflated in relation to other forms of artificial intelligence, with a particular focus on the content and forms of natural language used. One hundred and ninety-nine residents of the United States were recruited online and presented with a series of general knowledge questions. For each question they also received a recommendation from either an LLM or a non-LLM AI-assistant. The accuracy of this recommendation was also varied. All data is deidentified and there is no missing data.
This deidentified data may be used by researchers for the purposes of verifying published results or advancing other research on this topic.