What Does a Large Language Model Know?
- Featured in:
- Presidential Special Lectures
This Presidential Special Lecture was recorded Oct. 5, 2024, at Neuroscience 2024 in Chicago.
Recent developments in machine intelligence suggest the prospect of minds that can be artificially constructed. This raises a question: could an agent with such an artificial mind be intelligent, and if so, could it know things? In this lecture, Laurie A. Paul, PhD, raises this question with respect to large language models (LLMs) and suggest that an LLM can be thought of as realizing a certain type of mind, along with a certain type of knowledge. The lecture then explores the way that people can infer the existence of such knowledge and how it is related to the more ordinary sort of knowledge exhibited by human beings.