Meaning Matters for Large Language Models

Large language models (LLMs) have achieved remarkable adoption. While AI providers position their systems as assistive helpers and knowledge tools, users increasingly employ them for open-ended interactions seeking life-advice or creative exploration. This raises questions about the role of LLMs in such meaning-making activities, and the extend to which LLMs can access and encode meaning. In this conceptual essay, we apply Paul Ricoeur's hermeneutic philosophy to distinguish between structural and existential forms of meaning, revealing that LLMs can function as sophisticated conversational partners capable of engaging their vast “text” while lacking access to experientially-grounded understanding. We come to interpret user prompting as genuine hermeneutic encounters that enable meaning-making. We further reveal that hallucinations, the propensity of LLMs to generate plausible sounding yet incorrect responses, represent inevitable architectural trade-offs rather than eliminable technical failures. Our framework suggests new directions for LLM design that embrace generative capabilities and establishes principles for responsible user engagement.