> This one is tricky. We know exactly what they do. Interpreting that is very hard though, they've a big pile of mathematical operations with billions of magica

> This one is tricky. We know exactly what they do. Interpreting that is very hard though, they've a big pile of mathematical operations with billions of magical constants and it... works. We can see exactly what they do but if I could see every synapse firing in your brain I'd still not be able to understand how it works in a useful manner. So we understand them obviously, but at another level we really don't. Yeah but my bet is these numbers can all be traced to the training data, even when they show out-of-the-box thinking (like alpha go did with a move we didn't get at first). That it would be expensive and impractical doesn't mean we don't know what it does, and those claiming so are just trying to build hype. To those I have a few suggestions, build a simple visualization of the path it takes when it gives a surprising answer and it will make total sense. It's not practical to count every star in the galaxy either but we don't claim we don't understand them and selling it as magic (as some do). We find ways to make approximations we're happy with. > Just to be clear here, there is no recursion other than when you add more text. There is not an algorithm saying "identify parts X, then look in database Y, now summarise...". They're trained essentially to just predict the next word given some text. There's some later training to make them more conversational. The capabilities you see are just a consequence of that. This is contradictory. They ARE predicting the next word (really number, words are just our translation), then the next, then the next when building the response. Always in the context of what it's building, that IS recursion. And regarding "parts X, database Y", parts X is the prompt, find the vectors closest to that prompt in database Y, now build your best response. So it is working like you claim it isn't, but there are more steps. It is actually very simple in terms of database queries, the "magic" lies in the training and quality of the data. Genius in simplicity really and I'm not very surprised how well it works. To understand where I'm coming from I'm not sure we're even that intelligent and making conscious decisions as opposed to just reacting to stimuli. On a macro level we're very predictable and gullible and honestly disappointing. I'm trying very hard to keep my optimism and see our potential to be better but it's getting harder every year. > Perhaps instead you could give me a short question and the kind of answer that would surprise you? I'm not intelligent enough to come up with such a question and since I don't have access to the training data I can never verify that it's not just a discussion being reworded. An AI trained on a specific task starting to do a completely different task would be very cool and difficult to tell if sentient or not. That's why I instead focus on the mistakes that I spot, rather than when it feels more intelligent than myself. The pound of X vs Y is no longer a good question since they fixed it. I found that mistake here on HN, and it was obvious what it was trying to do. They are not stupid, but they are not sentient with agency either. With my level of cynicism setting the bar at "as good as humans" is a very low one. I know animals that impress me more. I want them to replace us, that should be the goal for our species. The problem with that is humans, we're still going to want infinite growth. Just look at the discussions being focused around what the unemployed are going to do after AI, not how we should put it to work for us so we can retire at birth. Gotta keep the wheels spinning, earth will probably multiply sooner or later. If not we'll just terraform Mars, parasites that we are. > I know this thread has gone on some time, but personally this is interesting to me. If you wanted to shift off from hn, feel free to drop me an email, I have a vested interest in understanding how people view LLMs. I agree it's been interesting, but I don't know if I'm a good source for how people view things. My views are more often than not regarded as controversial and strange.