I see a lot of people saying "it doesn't think" and while that's obviously true, I think it's important to expand on it a little. Not only is it the case that these models don't think in the classical sense of having sentience, they literally don't have any associations of meaning to the words. They don't put words together to express any semblance of meaning or some fact or anything like that. They are simply statistical engines that can link words together by their likelihood of appearing in some proximity of one another in their training data. It's not quite that simple because it doesn't see words the way we see words, it sees tokens which are transformed versions of the words that do carry some level of "meaning" but nowhere near the complexity of meaning in human speech. But there's a reason you can get Chat GPT to contradict itself very easily, the reason being that it doesn't assign meaning to what it was saying in the first place.