Eiichiro Kazumori (collaboration/inputs with UMass Dartmouth students)

(Source: https://phdcomics.com/comics.php?f=1583)

The bottom line of the state university is education and community. Given the crucial significance of education in the age of AI, it is critical to ensure this bottom line. Nevertheless, students, instructors, and the university currently spend a significant amount of time inefficiently on issues other than actual teaching and learning. Students need to spend significant time learning about student life, academic resources, technical issues, and campus information by looking at the university webpage, which is not necessarily easy to navigate. Often, students get lost before asking for help. At the same time, instructors and the university repeatedly spend a significant amount of time answering similar questions from students each semester. The UMass Dartmouth Large Language Model Helper automates the process to reduce inefficiencies, helping students get answers quickly so that students, instructors, and the university can refocus on education and community.

One use case is a student wanting to learn about the University of Massachusetts Dartmouth. A student may type the question, ‘Tell me about the University of Massachusetts Dartmouth.’ The chatbot will then provide an answer.

Further use cases (not implemented yet): (1) Students, instructors, and staff can check whether a facility is open without calling the facility. (2) Students can access course information from the syllabus without contacting instructors each time. (3) Students need not contact instructors and staff for tech support. (4) Students can learn class offerings that fit their needs and preferences. (5) Students can receive initial health support without going to the health center. (6). Students can obtain community contacts. 

The project will help students, instructors, and the university refocus on education and community, which is the bottom line of the state university.

Kazumori would like to thank the students who participated in the initial discussions that inspired this project when he discussed the generative TA project. Kazumori did all the work on his own. The current presentation of the project is preliminary, ongoing, and is subject to further revision.

We developed the first version in May 2024. The July 2024 version included data from UMass Dartmouth, WHOI, City of New Bedford, Town of Dartmouth, Town of Westport, and Southcoast Health. Due to current resource limitations as we teach classes, the current system has very limited capabilities. The chatbot interface is currently in reconstruction. In the long-run, we plan to implement novel algorithms to reduce hallucinations and improve response accuracy compared with current large language models, building on research described in Kazumori (2024).