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

AGI is far from inevitable

ru.nl

Post reply on HN
Print view View original

51–60 of 258 posts

Re: AGI is far from inevitable

#51

Post by graypegg »

I think the best argument I have against AGI's inevitability, is the fact it's not required for ML tools to be useful. Very few things are improved with a generalist behind the wheel. "AGI" has sci-fi vibes around it, which I think where most of the fascination is.

"ML getting better" doesn't *have to* mean further anthroaormorphization of computers, especially if say, your AI driven car is not significantly improved by describing how many times the letter s appears in strawberry or being able to write a poem. If a custom model/smaller model does equal or even a little worse on a specific target task, but has MUCH lower running costs and much lower risk of abuse, then that'll be the future.

I can totally see a world where anything in the general category of "AI" becomes more and more boring, up to a point where we forget that they're non-deterministic programs. That's kind of AGI? They aren't all generalists, and the few generalist "AGI-esque" tools people interact with on a day to day basis will most likely be intentionally underpowered for cost reasons. But it's still probably discussed like "the little people in the machine". Which is good enough.

Re: AGI is far from inevitable

#52

Post by jofla_net »

vundercind wrote: post #42
danaris wrote: post #35

Earlier quoted context omitted.

I have seen far, far too many people say things along the lines of "Sure, LLMs currently don't seem to be good at [thing LLMs are, at least as of now, fundamentally incapable of ], but hey, some people are pretty bad at that sometimes too!" It demonstrates such a complete misunderstanding of the basic nature of the problem that I am left baffled that some of these people claim to actually be in the machine-learning f…

I think the fact that this particular fuzzy statistical analysis tool takes human language as input, and outputs more human language, is really dazzling some folks I’d not have expected to be dazzled by it. That is quickly becoming the most surprising part of this entire development, to me.

At the very least, the last few years have laid bare some of the notions of what it takes, technically, to reconstruct certain chains of dialog, and how those chains are regarded completely differently as evidence for or against any and all intelligence it does or may take to conjure them.

Re: AGI is far from inevitable

#53

Post by godelski »

danaris wrote: post #35
godelski wrote: post #14

Earlier quoted context omitted.

> but LLMs do solve problems that people thought were extremely difficult to solve ten years ago. Well for something to be G or I you need them to solve novel problems. These things have interested most of the Internet and I've yet to see a "reasoning" disentangle memorization from reasoning. Memorization doesn't mean they aren't useful (not sure why this was ever conflated since... Computers are useful...), but it's…

I have seen far, far too many people say things along the lines of "Sure, LLMs currently don't seem to be good at [thing LLMs are, at least as of now, fundamentally incapable of ], but hey, some people are pretty bad at that sometimes too!" It demonstrates such a complete misunderstanding of the basic nature of the problem that I am left baffled that some of these people claim to actually be in the machine-learning f…

  > I have seen far, far too many people say
It is perplexing. I've jokingly called it "proof of intelligence by (self) incompetence".

I suspect that much of this is related to an overfitting of metrics within our own society. Such as leetcode or standardized exams. They're useful tools but only if you know what they actually measure and don't confuse the fact that they're a proxy.

I also have a hard time convincing people about the duck argument in [0].

Oddly enough, I have far more difficulties having these discussions with computer scientists. It's what I'm doing my PhD in (ABD) but my undergrad was physics. After teaching a bit I think in part it is because in the hard sciences these differences get drilled into you when you do labs. Not always, but much more often. I see less of this type of conversation in CS and data science programs, where there is often a belief that there is a well defined and precise answer (always seemed odd to me since there's many ways you can write the same algorithm).

Re: AGI is far from inevitable

#54

Post by tptacek »

Gehinnn wrote: post #18
tptacek wrote: post #15

Earlier quoted context omitted.

Are you talking about the press release that the story on HN currently links to, or the paper that press release is about? The paper (I'm not vouching for it; I just skimmed it) appears to reduce AGI to a theoretical computational model, and then supplies a proof that it's not solvable in polynomial time.

I was referring to the press release article. I also looked at the paper now, and to me their presented proof looked more like a technicality than a new insight. If it's not solvable in polynomial time, how did nature solve it in a couple of million years?

Probably by not modeling it as a discrete computational problem? Either way: the logic of the paper is not the logic of the summary of the press release you provided.

Re: AGI is far from inevitable

#55

Post by KoolKat23 »

jltsiren wrote: post #47
KoolKat23 wrote: post #26

Earlier quoted context omitted.

This I'm very sure will be the case, but everyone will still move the goalposts and look past the fact that different humans have different strengths and weaknesses too. A tone deaf human for instance.

There is another term for moving the goalposts: ruling out a hypothesis. Science is, especially in the Popperian sense, all about moving the goalposts. One plausible hypothesis is that fixed neural networks cannot be general intelligences, because their capabilities are permanently limited by what they currently are. A general intelligence needs the ability to learn from experience. Training and inference should not…

If that's the case, would you say we're not generally intelligent as future humans tend to be more intelligent?

That's just a timescale issue, if its learned experience of gpt4 is being fed into the model on training gpt5, then gptx (i.e. including all of them) can be said to be a general intelligence. Alien life one may say.

Re: AGI is far from inevitable

#56

Post by yourapostasy »

Peter Watts in Blindsight [1] puts forth a strong argument that self-aware cognition as we understand it is not necessarily required for what we ascribe to "intelligent" behavior. Thomas Metzinger contributed a lot to Watt's musings in Blindsight.

Even today, large proportions of unsophisticated and uninformed members of our planet's human population (like various aboriginal tribal members still living a pre-technological lifestyle) when confronted with ChatGPT's Advanced Voice Option will likely readily say it passes the Turing Test. With the range of embedded data, they may well say ChatGPT is "more intelligent" than they are. However, a modern era person armed with ChapGPT on a robust device with unlimited power but nothing else likely will perish in short order trying to live off the land of those same aborigines, who possess far more intelligence for their contextual landscape.

If Metzinger and Watts are correct in their observations, then even if LLM's do not lead directly or indirectly to AGI, we can still get ferociously useful "intelligent" behaviors out of them, and be glad of it, even if it cannot (yet?) materially help us survive if we're dropped in the middle of the Amazon.

Personally in my loosely-held opinion, the authors' assertion that "the ability to observe, learn and gain new insight, is incredibly hard to replicate through AI on the scale that it occurs in the human brain" relies upon the foundational assumption that the process of "observe, learn and gain new insight" is based upon some mechanism other than the kind of encoding of data LLM's use, and I'm not familiar with any extant cognitive science research literature that conclusively shows that (citations welcome). For all we know, what we have with LLM's today is a necessary but not sufficient component supplying the "raw data" to a future system that produces the same kinds of insight, where variant timescales, emotions, experiences and so on bend the pure statistical token generation today. I'm baffled by the absolutism.

[1] https://rifters.com/real/Blindsight.htm#Notes

Re: AGI is far from inevitable

#57

Post by ivanrooij »

The short post is a press release. Here is the full paper: https://link.springer.com/article/10.1007/s42113-024-00217-5

Note: the paper grants computationalism and even tractability of cognition, and shows that nevertheless there cannot exist any tractable method for producing AGI by training on human data.

Re: AGI is far from inevitable

#58

Post by nerdbert »

Gehinnn wrote: post #12

I skimmed through the paper and couldn't make much sense of it. In particular, I don't understand how their results don't imply that human-level intelligence can't exist. After all, earth could be understood as solar powered super computer, that took a couple of million years to produce humanity.

> In particular, I don't understand how their results don't imply that human-level intelligence can't exist.

I don't think that's what it said. It said that it wouldn't happen from "machine learning". There are other ways it could come about.

Re: AGI is far from inevitable

#59

Post by Atmael »

the point is that agi may already exist and work with you and your environment

you just won't notice the existence of agi

there will be no press coverage of agi

the technology will just be exploited by those who have the technology

Re: AGI is far from inevitable

#60

Post by klyrs »

The funny thing about me is that I'm down on GPTs and find their fanbase to be utterly cringe, but I fully believe that AGI is inevitable barring societal collapse. But then, my money's on societal collapse these days.
Post reply on HN