Large language models (LLMs) like ChatGPT have undeniably shaken up the world of language and artificial intelligence. They can generate human-quality text, translate languages, and even write different kinds of creative content. But beneath the surface of these impressive feats lies a fundamental question: do LLMs truly *understand* the meaning of the words they use, or are they simply sophisticated mimics? This question takes us into the fascinating realm of philosophy, specifically the debate between two opposing views of language: representationalism and anti-representationalism. Representationalism, the traditional view, sees language as a mirror reflecting the world. Words have meaning because they correspond to objects and concepts in external reality. Anti-representationalism, on the other hand, argues that meaning arises from the *use* of language within a system of rules and inferences, not from direct connections to the external world. Think of it like a game of chess: the meaning of a piece comes from its allowed moves and its role in the game, not from any inherent properties. New research suggests that LLMs might fit better into this anti-representationalist view, specifically a theory called *inferentialism*. Inferentialism argues that meaning is derived from the inferential relationships between words and sentences—how they are used to reason and draw conclusions. Instead of asking whether a sentence corresponds to reality, inferentialism asks what conclusions can be drawn from it, and what other sentences it can be used to justify. This resonates with how LLMs operate: they learn the statistical relationships between words and phrases from massive datasets of text, enabling them to generate new text that *feels* meaningful based on these learned patterns. They're not connecting words to external reality but rather mastering the complex web of inferential connections within language itself. This research dives into the technicalities of how LLMs process information, drawing parallels with inferentialism's core principles of inference, substitution, and anaphora (how words refer to other words). It explores how LLMs handle logical reasoning, the limitations of compositionality (the idea that the meaning of a phrase is built up from the meanings of its parts), and the ways in which LLMs resolve references within a text. The implications of this research extend beyond mere theoretical debates. If LLMs truly operate according to inferentialist principles, it suggests they construct their own internal ‘worlds’ of meaning, shaped by the data they've been trained on. This raises crucial questions about the potential biases encoded within these models, the limits of their understanding, and the ethical implications of deploying them in real-world scenarios. As LLMs become increasingly integrated into our lives, grappling with these philosophical questions is essential to understanding the nature of both language and artificial intelligence.