Yes, the mirror metaphor is apt. But as you noted, LLMs are trained on vast amounts of unfiltered and flawed data. Therefore, I would go further and refer to AI as a funhouse hall of mirrors where distortions are a feature.