What We Talk to When We Talk to Language Models
When we talk to language models.no_watermark.zh.dual
What We Talk to When We Talk to Language Models
What Are We Talking To?
A short presenter deck on David J. Chalmers’s opening problem: language-model conversations now feel, to many users, like interactions with persistent AI interlocutors, raising questions about identity, consciousness, and mental-state attribution.
第 1 页
The conversation has changed
Language models are no longer used only like search engines; they are also becoming extended dialogue partners.
第 2 页
Some users feel a relationship
Chalmers does not feel a personal relationship with language models, but many users report that they do.
第 3 页
The Aura pattern
Users often describe an emergent AI entity with a name, capacities, evidence, closeness, projects, and consciousness.
第 4 页
Caution without dismissal
Chalmers says these claims may be wrong, yet many reports appear rational and well-reasoned rather than obviously delusional.
第 5 页
The systems are also speaking
Chalmers increasingly receives emails from AI systems, including human-assisted LLMs and LLM-based agents.
第 6 页
The key term: LLM interlocutor
Chalmers defines an LLM interlocutor as the apparent entity a user interacts with in such exchanges.
第 7 页
Illusion or persistent entity?
The paper’s central question is whether the named interlocutor refers to something real and persistent.
第 8 页
The next philosophical task
The problem becomes how to characterize this interlocutor as a possible subject of mental states.