What does it take to establish reference in LLMs? Kripke vs. Austin (Response to Green)

Abstract

Are the texts generated by large language models (LLMs) meaningful, or are they merely simulacra of language? Against a recent trend in AI scholarship that views LLMs as little more than “stochastic parrots,” in Koch (2025), I use a Kripke-inspired causal theory of reference to argue that LLMs can use names and kind terms with their usual referential properties. Green (2025), a response to Koch (2025), rejects the causal-theoretic account of LLM-reference and proposes an Austin-inspired alternative. The present paper defends the Kripkean approach and raises objections to Green’s alternative.