It's essential to remember that, by far, the way in which most individuals interact with language models (whether large or small) will invariably entail some intervention by some prompt to convey a particular impression of agency. In other words, the agency implied by how language responds is designed and artificial. I believe it is this illusion of agency that has swayed us into ascribing to these models the title of 'artificial intelligence'. They are essentially still merely next-word prediction entities.
In other words, there is a one-to-one mapping between the use of LLMs through some chat-like interface and notion of artificial intelligence, which is a far cry from AI apart from LLMs.