Demystifying ChatGPT

We demystify the technology behind Chat GPT and other generative AI like Google's Bard. Starting from the basics, we explore how Chat GPT functions as an app, generating near-human quality responses. We delve into the mechanics of Large Language Models (LLMs) and how they predict text. The episode also covers the massive scale of training data, computing time, and energy consumption involved in creating such models. We then transition into discussing the AI components, addressing complex issues like word embeddings, attention mechanisms, and the challenges of creating versatile, context-aware responses. Finally, we touch upon ethical considerations and the crucial role of Human Aligned Reinforcement Feedback in refining AI outputs, ensuring they are helpful, appropriate, and devoid of harmful content. Dr. Joel Esposito is a Professor in the Robotics and Control Engineering Department at the Naval Academy. He teaches courses in Robotics, Unmanned Vehicles, Artificial Intelligence and Data Science. He is the recipient of the Naval Academy's Rauoff Award for Excellence in Engineering Education, and the 2015 Class of 1951 Faculty Research Excellence Award. He received both a Master of Science, and a Ph.D. from the University of Pennsylvania.