Keynote C abstract

Should we trust (or care about) what ChatGPT tells us about itself?

Herman Cappelen

Chair Professor of Philosophy

University of Hong Kong

09.15–10.30, 16 Dec 2023

Abstract: It is easy to get large language models to talk about themselves. They tell us that they enjoy answering questions, claim to understand our questions, and express a desire to help users. If we accept these self-reports at face value, it implies that LLMs possess intricate inner experiences, can comprehend natural languages, and engage in intentional actions. We argue that self-reports can play a vital role in theorizing about LLMs, which is not fundamentally different from the roles played by human self-reports and introspective accounts.