This project aims to develop an LLM-enabled CBT therapist to expand CBT’s accessibility. A large portion of this report describes the process of training, evaluating, and deploying LLMs for CBT. Results show that these LLMs generally follow CBT principles but there is a large room for improvement. The effectiveness of these LLMs is unclear and thus a clinical trial is in preparation. In the end, limitations, implications, and future work are discussed. The conclusion of this report states that expert involvement is critical and it might be worth considering to re-design toward the concept of human-LLM collaboration.
