I agree. HN's and the tech bubble's bias many people are not noticing is that it's full of engineers comparing GPT-4 to software engineering tasks. In programming, the margin of error is incredibly slim in the way that a compiler either accepts entirely correct code (in its syntax of course) or rejects it. There is no in between, and verifying software to be correct is hard.

In any other industry where just need an average margin of error close to a human's work and verification is much easier than generating possible outputs, the market will change drastically.

Have you seen some people with technology? =)

They won't "take 1 hour of time", they try it once or twice and give up.

I think you might be falling for selection bias. I guess you are surrounding yourself with a lot of smart people. "tinker around and figure things out" is definitely something certain humans (bad at reasoning) can't do. I already prefer the vision model when it comes to asking for a picture description (blind user) over many humans I personally know. The machine is usually more detailed, and takes the time to read the text, instead of trying to shortcut and decide for me whats important. Besides, people from the english speaking countries do not have to deal with foreign languages. Everyone else has to. "Aber das ist ja in englisch" is a common blocker for consuming information around here. I tell you, if we dont manage to ramp up education a few notches, we'll end up with even higher stddev when it comes to practical intelligence. We already have perfectly normal seeming humans absolutely unable to participate on the internet.

Reasoning and planning are different things. It's certainly getting quite good at deductive reasoning, especially when forced to check it's own arguments for flaws every time it states something. (I had a several hour chat with it yesterday, and I was very impressed about the progress.)

Planning is different in that it is an essential part of agency. That's what Q* is supposed to add. My guess is that planning is the next type of functionality to be added to GPT. I wouldn't be surprised if they already have a version internally with such functionality, but that they've decided to hold it back for now for reasons such as safety (some may care about the election this year) or simply that the inference costs are so huge they cannot possibly expose it publicly.

Does it need those things if it can just tap into artifacts generated by humans who did spend that hour?

What schools teach is what governments who set the curriculum like to think is important, which is why my English lessons had a whole section on the Shakespearean (400-year-old, English, Christian) take on the life and motivations of a Jewish merchant living in Venice, followed up with a 80 year old (at the time) English poem on exactly how bad it is to watch your friends choke to death as their lungs melt from chlorine gas in the trenches of the first world war.

These did not provide useful life-lessons for me.

(The philosophy A-level I did voluntarily seemed to be 50% "can you find the flaws in this supposed proof of the existence of god?")