It's not kindergarten child at congitive level. It's a different kind of cognition than humans, if you can call it "cognition" at all.

Regardless, it is foolish to belive it won't improve.

It's foolish to believe it will improve indefinitely or reach human parity.

Wasn't AI driving just few years away for 5 years?

The lesson I've taken from ChatGPT so far is that "consciousness" may be much less interesting or "special" than we thought. It may turn out that it's nothing more than an emergent property of a shared language that models the world around us and gives us capacity to plan and communicate our actions.

It's a word association machine, it's not even at kindergarten level when it comes to general cognition.

Regardless, it is foolish to belive it won't improve and take over tasks that in 2023 it still isn't able to do.

Who would guess, beyond Hitchhiker's Guide to the Galaxy, that I can use my phone in 2023 to translate anything on a foreign restaurant in real time.

It will of course continue to improve at what it does but I personally think it's unrealistic to assume it will somehow spontaneously develop generalised cognitive ability; there are surely limits to how far this particular approach will take us.

right now it will only solve problem someone else already knows the solution to, so not even at that.

that said, I expect an ai assisted clerk to be order magnitude faster than not. it will be though for people at the bottom of the learning curve for a bit, but in half generation the educational offering will include how to work toghether with ai, massively improving worker productivity

that is not to say it won't have negative impact. there's so much job that we need, currently programmer are in high demand, and it's the one of the highest paying jobs, but that will change, possibly dramatically. I expect people at the top of the chain to be in trouble first (architects and whatnot) because they are the least creative and the one that possibly require the most knowledge, things ai do exceptionally well as of today.

Having helped companies in traditional line of business to "streamline" their work processes, I am quite clear that many of the stuff that the West outsources into Asian countries will eventually be outsourced to AI.

It is already so that in many industries there are hardly any traditional coder jobs on site, and having to explain every step to avoid it going off rails in offshoring assignments is hardly going to be any different to explain it to an AI, when it gets good enough.

Maybe by GPT-20 only, but it will come, and then the roles of architects and business analists are the only ones left.

> Maybe by GPT-20 only, but it will come, and then the roles of architects and business analists are the only ones left.

Agreed. Although... we actually don't know which jobs will be the ones left. For all we know, it will have taken over business analysis, too. Plus, from my experience working on architecture, I suspect that automatizing that part of my job won't be too hard.

For all we know, the only jobs left will be nurse and deep sea miner.

There's no reason this couldn't happen, but surely it would require something better than LLM.

> right now it will only solve problem someone else already knows the solution to

This needs some explanation. What sort of novel problems can a person at kindergarten level solve, that GPT-4 cannot?

Aside from that, it's obvious that GPT-4 is finding solutions to previously unseen problems every minute of the day. What sort of criteria are you using to reject these as "someone else already knew the solution"?

It has no cognition. Do not personify the algorithm that cannot think.

It doesn't need to think per se, just like robots at an Assembly line, or algorithms at HFT, it only needs to do a good enough job.

Children don't have the same level of general knowledge in kindergarten. Adults don't have it either, not even the best adults.