Large Language Models (LLMs) like OpenAI’s ChatGPT or Google’s Gemini are the new sensation in artificial intelligence (AI) research. These systems exhibit impressive conversational abilities and have even managed to convince some people of their possible sentience. But do LLMs actually speak our language, or do they merely appear as if they do? Do they really reason and think, or are they simply good at superficially imitating these abilities? In this chapter, I argue that Wilfrid Sellars’s functionalist-pragmatist approach to language and concept learning might be especially useful in the context of answering the questions above. In particular, conceiving the process of learning and mastering language as analogous to the process of learning to play a game within a set of normative social practices can shed light on the kind of abilities LLMs possess and what we can expect them to do in the future, including becoming genuine members of our linguistic community rather than mere “stochastic parrots.”