The Data Revolution in Higher Education

GPT Is an Unreliable Information Store

11/08/2023

2:00pm - 2:50pm

Noble Ackerson

CTO

American Board of Design and Research

In this talk, we'll walk through how to make large language models more truthful and thus trustworthy at an enterprise scale using few or zero-shot approaches with a vector database, context injection, prompt chaining, and transfer learning. Why? Because large language models lack deductive reasoning or a cognitive architecture, which makes them epistemologically blind to what they know they know and their known unknowns. This makes them unreliable even with fine-tuning approaches alone.