If you look at the code generated by the latest and greatest models, you will still find it sloppy. So I won't say in general that models are already smarter than humans.
Plus the context windows are still so small, it's like the early days of computers where 256kb of memory was a lot. These little context windows don't let the models tackle very high level of abstraction of work. You still need to babysit them.
raunaksingwi
Distrust is definitely part of it, "smarter than almost everyone" isn't quite an accurate statement (it is more widely knowledgeable for sure, but "understanding" is still under-defined sometimes).
Lack of imagination is also part of it I'm sure... looking at the silly things some people ask AI (tbh so do I though, out of curiosity) but failing to use it for what's truly important and where it can add value (this is what I try to pursue, mostly by just trying it a lot on anything, without stopping to think for myself).
drltaidev
Personally, I don't think so.
The so-call human imagination is not always good and fantastic. The criteria of GOOD is very subjective. In most cases, some people don't know before the art shows. There are also some blind test about picking out AI-generated video and crafted one. No evidence tells the difference.
The worse part is AI is improved sooner and sooner while human beings not. I believe many will say they have no self-consciousness. What they generated are blind.
But doesn't really make sense if human cannot distinguish the source?
yancaoshi
Is taking advantage of Ai, related to financial gains? It is not the lack of imagination that stops people. They have to produce a product that is superior and market it and that requires capital and long term persistence. To top that off, more financially secure parties acquire patents and capital and monetize it. Zillions of people are looking for a way to capitalize on Ai and only a few get lucky.
silverFork
Why does change the world have to be an obligation? Lot of intelligent people I know are interested in other things.
hgu63
If you think AI is smarter than humans at "almost everything"...well, it might just be you.
On a side note, I really feel bad for Gen Z. Society has forced them to push AI as far as possible. They probably are the most aware of its limitations. Yet they are also the least able to criticize it without career repercussions.
I suspect this will change soon as more people come around to the fact that their chat bot is another tool just like Google search.
I used to think how wrong it was to compare an LLM to a search engine but have realized over time that despite being wrong, it's still a better mental model because it gets a fundamental detail right; LLMs are tools, they aren't agents, they aren't autonomous, they don't understand things, at least not in the way a human would understand them.
And this all ties back to the OP, that's the problem with calling them "smarter". Yes they can solve ancient math problems that humans failed at for many years and yet at the same time cannot be trusted to take a drive through order.
We need to stop seeing LLMs as a replacement for human cognition and instead appreciate what each form of "intelligence" brings to the table.
shahbaby
I think it may be capital gatekeeping more than bottlenecked intelligence. It feels like there is too large a gap between where people see and feel real problems firsthand and who can get the capital to fix it from either private or public sources
dv_dt
I believe it is due to the huge inertia that is the product of human nature. AI is advancing incomprehensibly fast, people aren't.
BSPtree
Change the world how?
I was one of the very first users and developers of Bitcoin, April 2009 - Dec 2015. So for better or worse I did take part in something that changed the world. And here's the thing about that: you don't want lots of people changing the world. The world is as it is for reasons and many of the people who attempted to change the world ended up downgrading it, if not for themselves then for at least some other people. Textbook example: the long history of utopian socialist revolutions that always ended up bathed in blood.
There are relatively few world changes nowadays that everyone would agree on. That's why the utopian pitch for AI is always the same tiny handful of ideas, like curing cancer. Beyond basics like curing the last diseases, there aren't a whole lot of changes you can make that are obviously positive because most problems faced by modern societies are purely socio-political problems. They aren't problems that yield to application of intelligence. They're all about clashes of values and people struggling to impose their will on each other.
Of the few where it seems intelligence might make a difference, there are mundane reasons why it doesn't help much. Medicine is bottlenecked on clinical trials. They have far more ideas than they can check already, so generating even more won't help. Physics is bottlenecked on particle accelerator power. Ain't no LLM going to fix that. What might help is figuring out how those ideas would work out ahead of time, but that's hard because we can't simulate the body or the universe and LLMs don't have any magical solution to that either.
So the biggest impact of AI will continue to be in software, a domain where we actually are intelligence bottlenecked. LLMs can systematically drain the swampy backlogs of a gazillion badly maintained projects and that could probably absorb all the datacenter capacity being built today, if it were cheap enough. They can let us try out new platform architectures and then rewrite all the software that sits on top. Etc etc. And yeah right now some of that is limited by our imaginations and some of it is limited by cost, speed and other things that will hopefully continue to improve post-bubble.
comments (9)
raunaksingwi
Lack of imagination is also part of it I'm sure... looking at the silly things some people ask AI (tbh so do I though, out of curiosity) but failing to use it for what's truly important and where it can add value (this is what I try to pursue, mostly by just trying it a lot on anything, without stopping to think for myself).
drltaidev
yancaoshi
silverFork
hgu63
On a side note, I really feel bad for Gen Z. Society has forced them to push AI as far as possible. They probably are the most aware of its limitations. Yet they are also the least able to criticize it without career repercussions.
I suspect this will change soon as more people come around to the fact that their chat bot is another tool just like Google search.
I used to think how wrong it was to compare an LLM to a search engine but have realized over time that despite being wrong, it's still a better mental model because it gets a fundamental detail right; LLMs are tools, they aren't agents, they aren't autonomous, they don't understand things, at least not in the way a human would understand them.
And this all ties back to the OP, that's the problem with calling them "smarter". Yes they can solve ancient math problems that humans failed at for many years and yet at the same time cannot be trusted to take a drive through order.
We need to stop seeing LLMs as a replacement for human cognition and instead appreciate what each form of "intelligence" brings to the table.
shahbaby
dv_dt
BSPtree
I was one of the very first users and developers of Bitcoin, April 2009 - Dec 2015. So for better or worse I did take part in something that changed the world. And here's the thing about that: you don't want lots of people changing the world. The world is as it is for reasons and many of the people who attempted to change the world ended up downgrading it, if not for themselves then for at least some other people. Textbook example: the long history of utopian socialist revolutions that always ended up bathed in blood.
There are relatively few world changes nowadays that everyone would agree on. That's why the utopian pitch for AI is always the same tiny handful of ideas, like curing cancer. Beyond basics like curing the last diseases, there aren't a whole lot of changes you can make that are obviously positive because most problems faced by modern societies are purely socio-political problems. They aren't problems that yield to application of intelligence. They're all about clashes of values and people struggling to impose their will on each other.
Of the few where it seems intelligence might make a difference, there are mundane reasons why it doesn't help much. Medicine is bottlenecked on clinical trials. They have far more ideas than they can check already, so generating even more won't help. Physics is bottlenecked on particle accelerator power. Ain't no LLM going to fix that. What might help is figuring out how those ideas would work out ahead of time, but that's hard because we can't simulate the body or the universe and LLMs don't have any magical solution to that either.
So the biggest impact of AI will continue to be in software, a domain where we actually are intelligence bottlenecked. LLMs can systematically drain the swampy backlogs of a gazillion badly maintained projects and that could probably absorb all the datacenter capacity being built today, if it were cheap enough. They can let us try out new platform architectures and then rewrite all the software that sits on top. Etc etc. And yeah right now some of that is limited by our imaginations and some of it is limited by cost, speed and other things that will hopefully continue to improve post-bubble.
mike_hearn