Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AGI is an engineering problem, not a model training problem

vincirufus.com

Post reply on HN
Print view View original

421–430 of 442 posts

Re: AGI is an engineering problem, not a model training problem

#421

Post by danenania »

habinero wrote: post #419
danenania wrote: post #376

Earlier quoted context omitted.

They lack many abilities of simple life forms, but they can also do things like complex abstract reasoning, which only humans and LLMs can do.

They don't reason. They can generate an illusion of it through a statistical model. You don't gotta work hard to break the illusion, either. People really really really want to believe this thing and I do not understand why. I wish I did lol

Ok, it’s an “illusion” of reasoning. Doesn’t change that it got a gold medal in the IMO or helped me fix a race condition the other day.

Re: AGI is an engineering problem, not a model training problem

#422

Post by danenania »

habinero wrote: post #420
danenania wrote: post #230

Earlier quoted context omitted.

It’s fine to think that—many clearly do. But it would be more honest and productive imo if people would just say outright when they don’t think AGI is possible (or that AI can never be “real intelligence”) for religious reasons, rather than pretending there’s a rational basis.

That's not a good way to think about it. Plenty of things could theoretically exist that aren't possible and likely will never be possible. Like, sure, a Dyson sphere would solve our energy needs. We can't build one now and we almost certainly never will lol "AGI" is theoretically feasible, sure. Our brains are just matter. But they're also an insanely complex and complicated system that came out of a billion years o…

> But they're also an insanely complex and complicated system that came out of a billion years of evolution.

As are birds, yet we can still build airplanes.

Re: AGI is an engineering problem, not a model training problem

#423

Post by satisfice »

More to the point: AGI is a testing problem. And it’s the kind of testing problem that can only be solved by hiring incompetent or corrupt testers.

Since there is no possibility of an uncontroversial specification of intelligence, there is no possibility of an honest and competent tester coming to the conclusion that sufficient testing has been performed on a candidate AGI to certify it.

Similarly, you won’t get any security professional to say that system is guaranteed to be secure.

Moreover, you can’t get any honest psychologist to swear that any person definitely has no mental illness.

How companies are dealing with this is to bet that they can fool enough people so that the remaining skeptics can be safely ignored.

Re: AGI is an engineering problem, not a model training problem

#424

Post by bsenftner »

Timwi wrote: post #388
dragonwriter wrote: post #368

Earlier quoted context omitted.

So you are now claiming that individual humans are not general intelligences and the only natural general intelligence is humanity as a unit ?

I have not seen any evidence of either. We have no way of knowing if we are “definitely” a true general intelligence, whether as individuals or as a civilization. If there is a concept that we are fundamentally incapable of conceptualizing, we'll never know. On top of that, true general intelligence requires a capacity for unbounded growth. The human brain can't do that. Humanity as a civilization can technically do…

I honestly believe it is an issue of communications, and our civilization does not take the skill of effective communications seriously, so a huge amount of information lost is really the defining factor of our entire civilization. We are incredible problem solvers, that have to repeat discovery of solutions over and over again because we fail to communicate their solutions or situations.

Re: AGI is an engineering problem, not a model training problem

#425

Post by germandiago »

jibal wrote: post #343
germandiago wrote: post #233

Earlier quoted context omitted.

Well, you should show the proof that it is possible also, So it would be a draw. I really think it is not possible to get that from a machine. You can improve and do much fancier than now. But AGI would be something entirely different. It is a system that can do everything better than a human including creativity, which I believe it to be exclusively human as of now. It can combine, simulate and reason. But think out…

The burden of proof is on the person who makes a claim, especially an absolute existential claim like that. You have failed the burden of proof and of intellectual honesty. Over and out.

You make the claim that AGI can exist, right? I make the claim it cannot. Both are claims, try to not manipulate the language. I acknowledge I started this but there is not a position that is better than the other.

Re: AGI is an engineering problem, not a model training problem

#426

Post by tim333 »

habinero wrote: post #418
tim333 wrote: post #316

Earlier quoted context omitted.

They pass human intelligence tests like exams and IQ tests. If I ask chatgpt how to get rid of spiders I'm probably going to get further than the spiders would scheming to get rid of chatgpt.

And Clever Hans could pass a math exam "Some tests can be cheesed by a statistical model" is much less sexy and clickable than "my computer is sentient", but it's what's actually going on lol

Some fibs and goalpost shifting there. Hans couldn't and sentience wasn't mentioned.

Re: AGI is an engineering problem, not a model training problem

#427

Post by fao_ »

Uehreka wrote: post #397
fao_ wrote: post #357

Earlier quoted context omitted.

> It doesn't seem clear that there is necessarily any connection between consciousness and intelligence. If anything, LLMs are evidence of the opposite. This implies that LLMs are intelligent, and yet even the most advanced models are unable to solve very simple riddles that take humans only a few seconds, and are completely unable to reason around basic concepts that 3 year olds are able to. Many of them regurgitate…

I suspect that you haven’t really used them much, or at least in a while. You’re spouting a lot of 2023-era talking points.

> I suspect that you haven’t really used them much, or at least in a while. You’re spouting a lot of 2023-era talking points.

I tested them recently and was not impressed, quite frankly.

Re: AGI is an engineering problem, not a model training problem

#428

Post by jibal »

jibal wrote: post #174
signa11 wrote: post #83

won’t somebody please think about Mr. Godel, and the Incompleteness Theorem ?

They aren't relevant. Even if Penrose and Lucas were right (they aren't), a computational system can solve the vast majority of the problems we would want solved.

P.S. I don't know which argument is allegedly "a bait and switch argument" .... certainly not mine. Even if Penrose and Lucas were right (again, they aren't--Godel's Theorems are in PA and can be proved by existing machines), that would not be a barrier to achieving AGI because being able to "see" the validity Godel's theorems is not a necessary condition for being an AGI.

P.P.S. Oh, I just noticed that the response came from the same person who wrote "won’t somebody please think about Mr. Godel, and the Incompleteness Theorem ?"

I will simply observe that isn't any sort of argument.

Re: AGI is an engineering problem, not a model training problem

#429

Post by player1234 »

danenania wrote: post #422
habinero wrote: post #420

Earlier quoted context omitted.

That's not a good way to think about it. Plenty of things could theoretically exist that aren't possible and likely will never be possible. Like, sure, a Dyson sphere would solve our energy needs. We can't build one now and we almost certainly never will lol "AGI" is theoretically feasible, sure. Our brains are just matter. But they're also an insanely complex and complicated system that came out of a billion years o…

> But they're also an insanely complex and complicated system that came out of a billion years of evolution. As are birds, yet we can still build airplanes.

[flagged]

Re: AGI is an engineering problem, not a model training problem

#430

Post by danenania »

player1234 wrote: post #429
danenania wrote: post #422

Earlier quoted context omitted.

> But they're also an insanely complex and complicated system that came out of a billion years of evolution. As are birds, yet we can still build airplanes.

[flagged]

Sorry you got triggered. I know it can be an emotional topic for some people. I'll try to explain in a simple way.

We clearly are replicating at least some significant aspects of human intelligence via LLMs, despite biological complexity. So we obviously don't need a 100% complete understanding of the corresponding biology to build things which achieve similar goals.

In other words, we can (conceivably) figure out how intelligence works and how to produce it independently of figuring out exactly how the human brain produces intelligence, just like we learned the laws of aerodynamics well enough to build airplanes independently of understanding everything about the biology of birds.

Whether we will achieve this or not to the point of AGI is a separate engineering question. I'm only pointing out how flawed these lines of argument are.

Post reply on HN