Faithfulness of LLMs in mental health conversations

Abstract

Faithfulness is a critical aspect of effective mental health conversa- tions, particularly when Large Language Models (LLMs) are deployed as virtual therapists. While LLMs have shown great potential in facilitat- ing mental health discussions, they often struggle with maintaining high levels of faithfulness—relying on accurate and empathetic responses that align with therapeutic principles. This project aims to develop a robust pipeline designed to enhance the faithfulness of LLMs in mental health contexts, ensuring that their responses are both accurate and aligned with the needs of the patient. Through this work, we seek to improve the relia- bility and effectiveness of LLMs as therapeutic tools in mental health care.