This study investigates the suitability of large language models (LLMs) as replacements for mental health providers. The authors find that LLMs exhibit stigma towards certain mental health conditions and often respond inappropriately to sensitive situations, even with safety guidelines and training. The research highlights the potential harm of deploying LLMs as therapists and emphasizes the need for caution and further research.

One author is a psychiatrist, and the study acknowledges funding from Microsoft and the Toyota Research Institute.

This paper presents a well-structured and important investigation into the limitations of current LLMs in mental health applications. It combines a review of clinical guidelines with targeted experiments, highlighting significant concerns about stigma and inappropriate responses. While the study acknowledges its limitations in scope and methodology, the findings are valuable and contribute to a growing body of literature on the responsible development of AI for mental health.