Yes. But then it’s no longer an LLM. And it still runs into the issues of taste and preference which would have to be included in its training set. At some point we’re no longer talking about AIs, but androids that are built to simulate the complete range of human experiences.

And this was Searle’s original argument. You can’t build intelligence by building strictly symbolic machines (computers). You have to endow them with semantic content by some other means.