Automated Collaborative Problem Solving Through Dialogues Between Specialized Large Language Models

Automated Collaborative Problem-Solving through Dialogues between Specialized Large Language Models Dr. Sean Mondesire, School of Modeling Simulation and Training, the University of Central Florida (UCF) A person wearing glasses and a suit Description automatically generatedDr. Sean Mondesire is an Assistant Professor at the University of Central Florida's (UCF) School of Modeling, Simulation, and Training (SMST). He is a part of the Knights Digital Twin Initiative, directs the Human-centered Artificial Intelligence Laboratory (HAIL), and co-directs UCF’s Advanced Research Computing Center (ARCC) for high-performance computing. His research specialties are machine learning and big data analytics for real-time recommender systems and autonomous decision-making at scale. Abstract: This presentation examines the use of Large Language Models (LLMs) in dialogic self-play for complex problem-solving. By assigning specialized roles to each LLM, we explore their capacity to tackle logic puzzles and improve NPCs in serious gaming. We focus on the efficiency of these AI dialogues in discovering innovative solutions and research questions, highlighting minimal human-in-the-loop approaches. This method demonstrates significant time-saving and creative potential, offering new insights into autonomous AI collaboration and its impact on advancing AI research and applications.