I'm not convinced that ChatGPT is anything more than just a fad in the current AI hype cycle, slightly more useful than Google's current answer snippets, but no

I'm not convinced that ChatGPT is anything more than just a fad in the current AI hype cycle, slightly more useful than Google's current answer snippets, but no more. ChatGPT still gets basic facts wrong, because it doesn't "understand" the material, it just relies on statistical correlations in language. But as philosophers and linguists have long known, outward forms of language are NOT the same as "understanding" the underlying concepts, though it can get tantalizingly close in appearance. In fact the impossible gap between language and logic has long been a favourite topic of both Eastern and Western philosophers. Language is hopelessly bad at expressing logical concepts unambiguously and correctly (hence the need for mathematical and logical symbols), and the only reason language works for us is because our brains are amazingly good at picking out the "relevant" meaning of sentences from literally thousands of possible interpretations. This happens at every level of language, from grammar and parsing sentence structure, to high-level conceptualization. But this is not a problem that can (according to current knowledge) be fixed with any deterministic or statistical model, and has been studied intensely for centuries. And so, this problem with ChatGPT's inaccuracy is not a problem that will be fixed with time. Like self-driving cars, the first 80% of the problem is trivially easy, but the remaining 20% is impossible, even with tens of billions in funding poured into this problem over the last decade. And the problem is, without that remaining 20%, the technology is useless because it needs to be close to 100% reliable for it to be useful at all. How can you verify the correctness of "facts" that ChatGPT so convincingly regurgitates? This is a recursive problem. Solving this problem would be tantamount to solving the hard problem of consciousness. We are still not anywhere close to Artificial General Intelligence, and one should attempt to at least understand the problems underlying this before thinking that ChatGPT is actually intelligent. See for example, John Vervaeke's study of Relevance Realization and "the impossibility of a general learning algorithm solution" [1]. [1] http://www.ipsi.utoronto.ca/sdis/Relevance-Published.pdf http://www.ipsi.utoronto.ca/sdis/Relevance-Published.pdf