The Wesley C. Salmon Memorial Lecture - Do Large Language Models Understand What They Are Talking About?

According to certain liberal accounts of meaning, the words produced by a large language model represent the things that they appear to represent. LLMs, in that case, make meaningful statements. But do they understand what they say? This paper uses contemporary accounts of understanding to explore the prospects for a positive answer to the question. A significant obstacle to understanding is LLMs’ isolation from the world that their words (arguably) depict. Nevertheless, an account of what it is for human thinkers to understand esoteric terms in scientific theories holds out the promise that current or future LLMs are capable of some degree of understanding in an interesting and substantive sense.