Focused Tutors: Assigning Custom, Targeted Chatbots to Students

Large language models have great potential to be pedagogical tools, able to interact with students as a tutor would, answering questions, addressing misunderstandings, using the Socratic method, etc. They can supplement in-person teaching with an availability and scale that cannot be economically provided by instructors and TAs. However, simply directing a generally available chatbot to be a tutor on a specific topic is likely to go down paths that are not relevant or appropriate for a specific class context. We have developed a ''focused tutor'' system in which instructors can create custom, targeted tutors by defining learning objectives and assessment questions. Instructors can deploy multiple tutors in a course, each tailored to a different module or goal in the specific context of that class, and assign them to students as low-stakes assessments following a reading or video. The intent is to provide a step between encountering material outside of class and working with it in class that is more engaging and provides more benefit to the students than a static reading response or reading quiz. In this lightning talk, we will present the motivation, design, and use of this tool, seeking feedback, ideas, and potential collaborations in studying this type of tool.