I have no idea about the future progression of AI, and anyone who claims they do is either lying to you or to themselves. I'll also mention: - ChatGPT was a l

I have no idea about the future progression of AI, and anyone who claims they do is either lying to you or to themselves. I'll also mention: - ChatGPT was a lot smarter for the types of things I'd like to use it for before the "safety" work. The efforts to make it never say anything offensive also made it a lot dumber. I saw a paper to that effect too. Perhaps some of that was a result of optimizations to shrink the model and make it more efficient (but then I'd expect GPT-4 to be smarter). - GPT3, which was an autocomplete, could do some things well the modern chatbot-ified models can't. - I have a lot of practice making the models smarter by having them check their own work, vote, etc. I have no idea how high the ceiling is. I might be a few inches away, or it might be the sky. That's a good time to hedge. Probably "legacy" was a better word than "obsolete." My language was sloppy.