A human mind probably is essentially a prediction engine, but it is predicting everything one can perceive, not just language. That's how a human can throw and catch a ball, or walk over uneven ground, or know that Mom will be mad about the mess.
Our intelligence is trained by hard reality, and so our use of language fits into a much broader context. "Touching a hot stove" is a metaphor because we know viscerally what would happen if we actually did it.
The limitation of ChatGPT is not that it predicts, the limitation is the incredible paucity of data from which it tries to predict. Language encodes some structure of the world because that's where it was created. But to the extent that can be inferred by ChatGPT, it's a weak and unreliable echo of the original reality.
ChatGPT exists in a featureless, timeless void with only patterns of characters in its memory. It can't tell real from fake (from our perspective) because ALL of its entire reality is just those patterns of text. The words are all real, and all there is.
This doesn't even get into why human minds make predictions. ChatGPT makes predictions because we built it to, and we prompt it to. Why do our brains? The trite answer is survival of the fittest, but that just raises the question of why we want to survive.
The point is, there is a fundamental difference between a living intelligence, and a machine learning tool. ChatGPT will sit quietly until prompted, and then answer only what it is asked. A human won't. I'm not sure that comes down to intelligence, or something else. Even bugs don't sit still and do what they're asked.