What We Talk to When We Talk to Language Models

What exactly are we talking to when we talk to language models such as ChatGPT, Claude, and Gemini? Do these models have genuine minds with mental states? I will argue that the LLMs we interact with are at least quasi-agents with quasi-beliefs and quasi-desires, in a sense I will explain. I will also argue that the LLM quasi-agents that we interact with are best understood not as abstract models or even as hardware instances but as virtual entities bound to conversation-based memory threads. I will draw some parallels with the TV show Severance.

Speaker: David Chalmers, New York University

Register at weblink