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

Rodney Brooks on GPT-4

spectrum.ieee.org

Post reply on HN
Print view View original

231–240 of 412 posts

Re: Rodney Brooks on GPT-4

#231

Post by whimsicalism »

fwlr wrote: post #33

> The large language models are a little surprising. I’ll give you that. I think this is the key point about LLMs that kind of explains the wide and polarized views on whether it understands or parrots, whether it can think or is the precursor to thinking or is a dead-end, whether it will catastrophically destroy the world, or “merely” make it steadily worse with bullshit, or just put a few industries out of a job. A…

> But almost everybody is really surprised that becoming better at a task like ‘natural language prediction’ would produce all these strange abilities that sort of look like “understanding the world”.

It honestly should not have been a surprise to anyone in the field at least in the last 6 years.

Re: Rodney Brooks on GPT-4

#232

Post by whimsicalism »

notShabu wrote: post #177
fwlr wrote: post #33

> The large language models are a little surprising. I’ll give you that. I think this is the key point about LLMs that kind of explains the wide and polarized views on whether it understands or parrots, whether it can think or is the precursor to thinking or is a dead-end, whether it will catastrophically destroy the world, or “merely” make it steadily worse with bullshit, or just put a few industries out of a job. A…

Has there been any research in the possibility that much of what we say/speak (including this convo) is actually just stochastic parroting? e.g. instead of the stochastic parrots mimicking intelligence maybe intelligence doesn't exist, it's just stochastic parrots of various levels of sophistications organized into a hierarchy. "Intelligence" is necessarily socially defined with the more complex parrots being unpredi…

> Has there been any research in the possibility that much of what we say/speak (including this convo) is actually just stochastic parroting?

I don't think that is really a well defined question. What is a stochastic parrot?

Re: Rodney Brooks on GPT-4

#233

Post by anon84873628 »

Sai_ wrote: post #181
notShabu wrote: post #177

Earlier quoted context omitted.

Has there been any research in the possibility that much of what we say/speak (including this convo) is actually just stochastic parroting? e.g. instead of the stochastic parrots mimicking intelligence maybe intelligence doesn't exist, it's just stochastic parrots of various levels of sophistications organized into a hierarchy. "Intelligence" is necessarily socially defined with the more complex parrots being unpredi…

This is an interesting question. To paraphrase as per my understanding of your comment, is intelligence an emergent property of being able to interact with each other through language? Say I speak gibberish (to you) which is actually me explaining to you the theory of relativity, would you consider me intelligent?

What's the difference between "understanding" and "having really good probabilistic information about how words combine"?

Kids learn to speak by parroting what they hear and observing the outcome. Then they run tests that reinforce the connections between words. That's what the model is.

But humans also get to link words with all the other sense experience we have (like how sweet cherries, loud fire trucks, and that one crayon are all "red"). LLMs don't have as many dimensions of experience they can link to.

But anyway, intelligence is about having an internal model of the world and using it to predict the future. The more rich and accurate the model, the more intelligent. The ability to communicate isn't a prerequisite; lots of animals have intelligence that isn't built with language.

Re: Rodney Brooks on GPT-4

#234

Post by whimsicalism »

akiselev wrote: post #161
yblu wrote: post #149

Earlier quoted context omitted.

What I found disheartening was many of those scientists, especially those on the "nothing to worry about" camp, seemed not to entertain the thought that they could be wrong, considering the scale of the matter, i.e. human extinction. If there's a chance AI poses an existential threat to us, even if it is 0.00000001% (I made that up), should they be at least a bit more humble? This is uncharted domain and I find it in…

Meh. Add it to the pile. The number of world ending risks that we could be worried about at this point are piling up and AI exterminating us is far from the top concern, especially when AI may be critical to solving many of the other problems that are. Wrong about nuclear proliferation and MAD game theory? Human extinction. Wrong about plasticizers and other endocrine disruptors, leading to a Children of Men scenario…

I think almost none of the scenarios you have named outside of the asteroid & the AGI would result in complete human extinction, potentially a very bad MAD breakdown could also lead to this but the research here is legitimately mixed.

Re: Rodney Brooks on GPT-4

#235

Post by whimsicalism »

onethought wrote: post #179
akiselev wrote: post #161

Earlier quoted context omitted.

Meh. Add it to the pile. The number of world ending risks that we could be worried about at this point are piling up and AI exterminating us is far from the top concern, especially when AI may be critical to solving many of the other problems that are. Wrong about nuclear proliferation and MAD game theory? Human extinction. Wrong about plasticizers and other endocrine disruptors, leading to a Children of Men scenario…

Yes, and all of those (including AI) are not even human extinction events. - Nuclear war: Northern Hemisphere is pretty fucked. But life goes one elsewhere. - Plasticisers: We have enough science to pretty much do what we like with fertility these days. So it's catastrophic but not extinction. - Climate Change: Life gets hard, but we can build livable habitats in space... pretty sure we can manage a harsh earth clima…

> - Astroid impact: Again, ALL human life globally? Some how birds survived the meteor that killed the dinosaurs, I'm sure we'd find a way.

I agree with many of these but we'd plausibly be toast in this scenario.

Re: Rodney Brooks on GPT-4

#236

Post by DJHenk »

zmmmmm wrote: post #169
JoshTko wrote: post #38

Earlier quoted context omitted.

The more I think about it the more I'm convinced I am basically just predicting/saying my next word whenever I speak.

Honestly I feel like one reason people are struggling with this is they can't accept a critical part of the truth : most people are stochastic parrots themselves most of the time. True, creative genuine deep thinking is an exceptional state of thinking for us.

Years ago I transitioned from a developer role to a manager role and suddenly I had to do a lot more talking. Not all the talking needs to be a deeply involved exchange of complex ideas, a lot of it serves a different purpose. Sometimes it can be a simple as filling up the time in a pleasant way with a group of people that may or may not know each other that well.

After getting some experience with this I noticed that I had developed a talking on/off button in my head. I could just simply turn it on and start talking. I could generate words that sounded good together and fit the purpose of the moment. But They just seemed to come from a different place in my brain than my conscious mind. Because that was not involved in this process at all. The only job my mind had was to turn the button off again at the right moment, for the rest it was free to think whatever it wanted.

(I transferred back to development a couple of years later.)

Re: Rodney Brooks on GPT-4

#237

Post by whimsicalism »

causality0 wrote: post #49
fwlr wrote: post #33

> The large language models are a little surprising. I’ll give you that. I think this is the key point about LLMs that kind of explains the wide and polarized views on whether it understands or parrots, whether it can think or is the precursor to thinking or is a dead-end, whether it will catastrophically destroy the world, or “merely” make it steadily worse with bullshit, or just put a few industries out of a job. A…

If LLMs are an understanding of the world it would mean humans in a few decades found a way to create sapience with many orders of magnitude fewer interacting elements than evolution did. I find that doubtful, at least in light of the fact every other way we've replicated biological computation requires many more computational elements.

If humans actually went to the moon, it would mean that humans in a few decades found a way to access a niche that nature never did.

In all seriousness, it's interesting all of these dualisms we like to hold on to. Humans are part of nature. It is unsurprising that further sapience would branch off from an already sapient race as opposed to re-emerge elsewhere.

Re: Rodney Brooks on GPT-4

#238

Post by mejutoco »

beardedwizard wrote: post #105
krainboltgreene wrote: post #90

Earlier quoted context omitted.

> it's "read" a significant fraction of everything we've ever written about chess I doubt heavily that a significant fraction of chess's writings are even available in digital format, much less inside of CommonCrawl and correctly trained on.

Why would common crawl NOT contain a chess manual? the rules are explained in detail on wikipedia. the simplest conclusion is that it has indeed been trained on a chess manual and is good at predicting what the next word in a chess manual is. it is not synthesizing anything.

"A chess manual" vs the finer point of OP "a significant fraction of chess's writings"

Re: Rodney Brooks on GPT-4

#239

Post by croes »

jiggawatts wrote: post #110

This is a terrible article written by someone who doesn't seem to have even tried GPT 4. Their only example references GPT 3.5, for example, and then they waffle on about only vaguely related topics such as level 5 self-driving. This quote in particular stood out as ignorant: “What the large language models are good at is saying what an answer should sound like, which is different from what an answer should be.” That…

>The elephant flew to the Moon.

Doesn't prove anything. So GPT-4 is trained on Wolframs example or many people tried it on GPT-4 and corrected the wrong answer.

Re: Rodney Brooks on GPT-4

#240

Post by ly3xqhl8g9 »

jmyeet wrote: post #70

GPT-4 is pretty amazing but I, too, feel this is being overhyped. For me, a sobering example is how OpenAI does math (eg [1]). Specifically, the model clearly doesn't really understand multiplication and "learns" it from training data. This tends to get the first few and last few digits right for a simple multiplication with 6-7 digit numbers. Now you can solve that with plugins (eg training the model to recognize ma…

Large language models and the transformer architecture are just the ALGOL 60 [1] of the search for general synthetic competence (intelligence is too finicky, competence is being given a task and fulfilling it successfully, in time, on budget, optimizing along the way). ChatGPT + Wolfram or other plugins are great for making transformers mathy, but the plugins must also be end-to-end machine learning architectures. To continue the high-level programming language analogy, we will probably see the C-level neural architecture in a few years once we get more embedded data (competence obtained through correlation physical object ↔ concept) with feedback loops forcing algorithms to operate under energy-restrictions (unlike the current architectures, more of a proof of concept, being careless about their energy).

[1] https://en.wikipedia.org/wiki/ALGOL_60

Post reply on HN