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

Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

news.ycombinator.com

Post reply on HN
Print view View original

51–60 of 62 posts

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#51

Post by grayhatter »

richk449 wrote: post #46
grayhatter wrote: post #17

Earlier quoted context omitted.

> s haven’t tried GPT-4 or they have and are just stubbornly refusing to change their mind about something when presented with new evidence. Have you considered that the evidence just isn't convincing yet? If you view popular llms as text prediction machines, they are in fact much better than spell check or auto correct from a few years ago. But if you actually ask it to solve the problem with nuance it will not use…

It’s amazing how easily people are fooled into thinking something is intelligent. I met Stephen Hawking once and asked him how to slice video with ffmpeg. He didn’t have a clue! Obviously he isn’t an intelligent person. Yet so many people think he is smart.

Your strawman argument isn't impressive either, even chatGPT knows how to avoid them! But I'll play along! Didn't happen to try something different, like playing 20 questions? Or something else? Because I've already given those as other examples for simple tasks that I don't think a banal text prediction algorithm could do.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#52

Post by arisAlexis »

xkapastel wrote: post #44
arisAlexis wrote: post #25

Earlier quoted context omitted.

But 2 out of 3 scientists that created the tech you are talking about disagree with you. Makes you wonder if..?

This is a strange way to count, the three Turing award winners are not the ones that "created the tech". Many thousands of people did that.

It's common to call them godfathers of AI. Surely Oppenheimer didn't create the bomb alone but you get the point.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#53

Post by arisAlexis »

qazwsxedchac wrote: post #30
arisAlexis wrote: post #24

Earlier quoted context omitted.

https://yoshuabengio.org/2023/05/22/how-rogue-ais-may-arise/

Thank you for the reference. Its author defines a "rogue AI" as: "an autonomous AI system that could behave in ways that would be catastrophically harmful to a large fraction of humans, potentially endangering our societies and even our species or the biosphere" and explains that it would also need to be goal-directed in a way which would be at odds with human wellbeing. Stipulating all that, what is still missing is…

The article explains that too. But it's easy to think of millions of examples if you can hack everyone and persuade people to do stuff for you and you are smarter than them.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#54

Post by hnfong »

grayhatter wrote: post #17
jstx1 wrote: post #14

Earlier quoted context omitted.

> But it's very rare when someone is impressed when the model is generating text based in something they do understand deeply. This statement sounds out of date. And you can see this sentiment a lot on HN. I don’t know if the people who say this haven’t tried GPT-4 or they have and are just stubbornly refusing to change their mind about something when presented with new evidence.

> s haven’t tried GPT-4 or they have and are just stubbornly refusing to change their mind about something when presented with new evidence. Have you considered that the evidence just isn't convincing yet? If you view popular llms as text prediction machines, they are in fact much better than spell check or auto correct from a few years ago. But if you actually ask it to solve the problem with nuance it will not use…

I think the issue is just that people are using different words to mean the same thing.

If we get precise, take for example what you said earlier:

> it's slightly better than the mean across all topics.

Most people who can do "better than average across all topics" usually aren't considered "dumb" at all. They might even be considered impressive, albeit being a jack of all trades might be a career disadvantage.

In these discussions I see (generally) a strong sentiment among the naysayers along the lines of "because ChatGPT is so overhyped, I've decided to play it cool and downplay it as a reaction to the hype". And hence the other comment suggesting that you're intentionally not allowing yourself to be impressed by the state of the art achievements.

IMHO there's too much arguing over how we should subjectively "feel" about the new tech. Everyone should be free to "feel" whatever they want to feel about the state of things, whether being impressed or not.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#56

Post by sharemywin »

mplewis wrote: post #32
sharemywin wrote: post #16

I think it's very hard to see what happens when these things can handle video and become truly multi-model. you would get a very powerful world model, I think. There don't seem to be any sensors you can't hook them up to. could it learn to infer left from right, object permanence, things like falling and gravity? Another thing to look at is large organizations accomplish much more than any individual in them, even th…

What things specifically? LLMs?

Not LLMs.

Here's a paper that's doing video prediction. https://paperswithcode.com/task/video-prediction

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#57

Post by lisasays »

vannevar wrote: post #48
lisasays wrote: post #41

Earlier quoted context omitted.

Even if the statement were true, the data that these models are trained on is, collectively, smarter than the smartest person who ever lived. "Smarter" in a horizontal sense. But not vertical. The latter being what people really mean by "smarter". You know, that thinking-outside-the-box stuff. That's why LLMs aren't able to "solve" problems that don't have existing writeups somewhere (beyond certain problems more or…

>That's why LLMs aren't able to "solve" problems that don't have existing writeups somewhere (beyond certain problems more or less equivalent to combinatorial search). Neither can humans. We create by analogy; "thinking outside the box" just means making less obvious (but still valid) analogies. >Or produce text useful beyond the mild amusement level (or for spam and/or research fraud -- at which it of course excels…

Neither can humans. We create by analogy;

This seems to be at (very, very) best unsubstantiated; most likely an insufficient explanation of how humans create.

People are using LLMs very effectively for writing news reports, opinion articles, ...

Well, we certainly differ in our assessments on that. Literally everything I've seen in the first two categories is complete garbage.

In the sense of being either patently unreliable (news), or simply having nothing interesting to say (opinion).

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#58

Post by xkapastel »

arisAlexis wrote: post #52
xkapastel wrote: post #44

Earlier quoted context omitted.

This is a strange way to count, the three Turing award winners are not the ones that "created the tech". Many thousands of people did that.

It's common to call them godfathers of AI. Surely Oppenheimer didn't create the bomb alone but you get the point.

The point is wrong, they don't have any claim to having created deep learning and so their perspective can't be used to argue for the reality of "superintelligence". Even if they did create it, the question of whether "superintelligence" is a meaningful concept or not isn't really related to what they would have had to go through in order to make deep learning work.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#59

Post by antifa »

richk449 wrote: post #47
qazwsxedchac wrote: post #21

> sober voices as Geoffrey Hinton sounding the alarm about super-smart AI Can someone please explain to me what exactly the danger is / the dangers are of "super-intelligent" AI? AIUI, an AI is a combination of hardware, software and parametrization. In broad terms, it exists as a black box which supplies to humans responses to token sets fed to it. Even if an AI has the launch codes for ICBMs somewhere in its traini…

> Even if an AI has the launch codes for ICBMs somewhere in its training data, it doesn't have an interface to the nearest missile silo to use them. It cannot commandeer the resources (hardware, space, cooling, electricity) it needs to operate, it is dependent on humans to supply those. How do you know the AI doesn’t have an interface to the middle silo? Are there no military systems that are connected to the interne…

> I'm not worried that gpt4 is going to start launching missiles

Cool SciFi showerthought:

An AI will probably know that Stuxnet happened.

https://en.m.wikipedia.org/wiki/Stuxnet

An AI could trick multiple human actors into unknowingly assembling all the pieces together for a nuclear launch.

Re: Ask HN: Why do so many assume we’re on the cusp of super-intelligent AI?

#60

Post by arisAlexis »

xkapastel wrote: post #58
arisAlexis wrote: post #52

Earlier quoted context omitted.

It's common to call them godfathers of AI. Surely Oppenheimer didn't create the bomb alone but you get the point.

The point is wrong, they don't have any claim to having created deep learning and so their perspective can't be used to argue for the reality of "superintelligence". Even if they did create it, the question of whether "superintelligence" is a meaningful concept or not isn't really related to what they would have had to go through in order to make deep learning work.

So in computer science nobody can claim that they are the godfather of X. In physics too because everyone was based on mathcfrom ancient Greece. I get your point. Let's focus on the important stuff. 2 out of 3 Turing award winners related to this tech put up huge warnings. You think a) ignore or b) take seriously?
Post reply on HN