In this one-hour session, we will delve into the capabilities and limitations of GPT-3 (the model at the core of ChatGPT) and its evolution from a “BS’ing machine” to a knowledge learner. Using a simple metaphor, we will explain how GPT-3 works and how it can be taught through retraining the model or integrating a corpus of “facts.” Using a couple of historical metaphors, we will offer some clues as to how the tools used in building large language models, such as content management systems, may evolve.

We will showcase several examples of GPT-3-based applications, including a historical society chatbot, a narrow domain expert chatbot, and the cognitive re-engineering of business processes, to illustrate the integration of “facts” into GPT-3. The architecture and findings of each example will be discussed.

Additionally, we will offer some prerequisites for conducting experiments and building applications within a domain using GPT-3, such as a machine-readable text corpus, a self-contained domain, and minimal resources such as Python and a small budget.

Towards the end of the session, we will engage in a group discussion about potential applications of GPT-3 in different fields and explore opportunities for hypothecated domains. Attendees will leave the talk with a deeper understanding of GPT-3 and its potential implications in various domains.

Please register. Speaker bio and registration at bit.ly/3ZYkyMf