Dialogic Implementation: The Group-in-the-Loop Large Language Model System

• In Event: 50462 - Section on Science, Knowledge, and Technology Roundtables
  In Refereed Roundtable (60 minute): Table 3: LLMs, a SKAT perspective

Abstract

Finetuning, reinforcement training, and other processes for improving large language model (LLM) performance focuses on improving the product. The presumption is that the individual users require a general-purpose knowledge tool for navigating some terrain of information. Understanding the nature of LLM’s is usually a technical question that needs to be disentangled from the modern systems that made the emergence of LLMs, and machine learning more generally, possible.The goal of this project is to develop a group-level process where LLM implementation occurs within a feedback loop collectively embodied by organic humans. Group-level dialogue between users becomes central to defining the epistemology, knowledge corpus, and performance of the LLM. This paper describes the “Hawk AI” project’s process of constructing a “Group-in-the-Loop” LLM system for diagnosing challenges and strategies for student wellness in higher education.

Authors

• Christopher Shane Elliott, UNCW

• Sang Teck Oh, University of North Carolina Wilmington

• Sathvik Thota, University of North Carolina Wilmington

• Luke Butler, University of North Carolina Wilmington

• Douglas J. Engelman, University of North Carolina Wilmington

• Christain Cole, University of North Carolina Wilmington