Hacker News Reader

Comment by codingdave | Hacker News Reader

I'm still so confused by this attitude. ChatGPT is not accurate. It cannot innovate, it only can rehash content is has been provided. Sure, it does so mind-boggling well. It appears to be masterful at implementing requests... but it does not know when it did something wrong. It cannot fact check itself. It cannot test its own work. And while it can reshuffle words into something new, it cannot actually invent anything new -- that depends on a person guiding it, judging the results, and driving it through additional iterations.

It is a great tool, but it is just a tool.

View on HN

Another case of moving the goal posts.

It used to be thought that only humans could play chess, as chess playing was seen as a mark of human intelligence, then when computers could play chess it was thought only humans could play chess well, and when computers could play chess well it was argued that computers couldn't beat the best of human chess players, so humans were ahead. But then computers started to beat the best hummer chess players. So the sights were next set on Go, which went through the same cycle, but much faster until computers could outplay the best human Go players.

The evolution of AI is littered with such hurdles that computers overcame time and time again, but the goal posts keep moving.

We should note that some humans can't do the things you ask for either... or not well, and if they can they don't always do so (most people speak in cliches, for instance, so could be said to be "just reshuffling words"), and, anyway, computers are quite capable of generating randomness, which is definitely "new" in some sense.

So what's the answer to the goal post problem, should we accept Deep Blue as being a fully human level AI intelligence? Maybe we should grant it citizenship. I'm not sure what conclusion you're alluding to but not stating.

It's unlikely politics will be able to control strong AI either. That may be where politics, and humanity as a whole, may finally meet its match.

Oh I agree, true strong AI that's actually beyond human cognitive capabilities, will by definition be beyond human control. You can't control something that is comprehensively beyond your own limitations, in the long term anyway. That's why the alignment problem in AI is an existential issue for us.

Even the primitive, weak AIs we have now are already way beyond human capabilites in many ways.

One they acquire the je ne sais quoi of consciousness, we're toast.

I just generated an httpkit clojure web app hat fetches coin and price info from coinmarketcap in 15 minutes. This would have been a minimum of 4 hours based on my current clojure ability. It's doesn't have to be 100% right. You are still expected to grok code. It's simply an assistant and it's bloody amazing.

Reply on news.ycombinator.com