Thank you for using the Practice Patients automated standardized patient for your learning experience! We hope that this can be a helpful learning tool for exploring symptoms, diagnosis, and patient dialogue.
This system works by using a process called Retrieval Augmented Generation (RAG). With RAG, a standard language model is provided with specific background information, and when you speak to the model it responds using information from this background. In this case, the background features information related to a fictionalized patient. Generally, this process seems to effectively capture some of the characteristics of a typical patient interview in a clinical setting.
Please see below for some helpful tips and information regarding your interactions with the system:
The automated patient will usually stick to their role as your patient. However, sometimes the model can get confused—for example, it may ask how it can help you. To get back on track, ask a question that you would typically ask a patient, and this will push the model toward behaving as the patient.
The model is designed to respond realistically. For example, if you ask the model to write a poem about its symptoms, the system will usually decline the invitation. However, this is not foolproof; unusual questions may get unusual answers. For a realistic patient-dialogue experience, we recommend that you ask questions that you would realistically ask a patient.
Conversations with PracticePatients models are recorded to understand and improve the system. These are 100% anonymous and can’t be tracked back to individual users. Nevertheless, keep in mind that dialogue with PracticePatients is recorded.
Questions, comments, or concerns can be directed to schulmar@upstate.edu.