That's a terribly narrow way to interpet music and art in general. It isn't just the mere mechanism of production that makes art so impactful to people - it's the dialogue and conversation back and forth in a medium that isn't always explicit in its intent. I mean yeah sometimes in rap, it's pretty clear what their message is. Kendrick out front shoulda told ya.
But with that example, what made Kendrick's music so interesting was its context and delivery and collaboration with other artists.
Hell, even just as a musician, you can take the same notes and rhythm on the page and produce a wildly different communicated emotion. Music and art is far more than just an idea, it's the execution and context its delivered that brings so much more meaning.
When the thing you're in conversation with is just a set of seven hundred billion parameters in a very very very very large parametric equation...... That's sort of missing the partner in the discussion that imbues art with its metaphysical meaning and value. Not to mention, LLM's definitionally produce an "average" of all the information they've been trained on. They definitionally cannot take in and internalize new information like a conversational partner can. It's very literally talking at a wall and you will never convince it of anything, until you spend another ten million bucks on a new training run for your large language model to give it a new set of ~700,000,000,000 parameters.
That's just my opinion at least, both as a life long musician and as a professional technical specialist in machine learning and large language models in particular (and a few other discliplines as well, dm me if you want to nerd out).