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

51–60 of 412 posts

Re: Rodney Brooks on GPT-4

#51

Post by idopmstuff »

> The example I used at the time was, I think it was a Google program labeling an image of people playing Frisbee in the park. And if a person says, “Oh, that’s a person playing Frisbee in the park,” you would assume you could ask him a question, like, “Can you eat a Frisbee?” And they would know, of course not; it’s made of plastic. You’d just expect they’d have that competence. That they would know the answer to the question, “Can you play Frisbee in a snowstorm? Or, how far can a person throw a Frisbee? Can they throw it 10 miles? Can they only throw it 10 centimeters?” You’d expect all that competence from that one piece of performance: a person saying, “That’s a picture of people playing Frisbee in the park.”

This seems like exactly a set of things that GPT-4 can do. The image recognition capabilities haven't been released yet, but they were demoed when it launched and clearly have the ability to handle a situation like this. From there, you could ask it every single one of these questions and get the correct answer.

Re: Rodney Brooks on GPT-4

#52

Post by astrange »

famouswaffles wrote: post #21

No world model ? A world model is so obvious papers like these are more confirmation than surprise https://arxiv.org/abs/2305.11169 https://arxiv.org/abs/2210.13382 There a certain sentiment that AGI however you wish to define it won't infact be a "We'll know it when we see it" situation but rather a "AGI will arrive long before consensus reaches its AGI". LLMs have made me believe this will 100% be the case, either…

Do you know what a "world model" is? It's a thing people were assumed to have in 1970s psychology, but was never well-defined enough to tell if it exists or not, so I don't think it's obvious anything else has one.

https://twitter.com/Meaningness/status/1639120720088408065

I think "common sense" or "long term memory" might be more productive things to say.

Re: Rodney Brooks on GPT-4

#53

Post by isignal »

runthemodel wrote: post #32

At least run the model on the examples considered... Here's GPT3.5 > Can you eat a Frisbee? No, you cannot eat a Frisbee. A Frisbee is typically made of plastic, often polypropylene or similar materials, which are not meant for consumption. These materials are not digestible and can pose a choking hazard or harm your digestive system if ingested. It's important to only consume food and items that are safe and intende…

The claim wasn’t that chat gpt doesn’t know what a frisbee is. The claim was that recognizing frisbees is different from a fundamental understanding of frisbees. So it was an example rather than a specific claim about chat gpt.

Re: Rodney Brooks on GPT-4

#55

Post by zht »

ramraj07 wrote: post #47
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…

The issue with all these experts is they still think it's human nature to be able to fully understand the world before they speak about it. On the contrary it's human nature (and all animal nature) to figure out how to navigate the world without fully understanding or having a complete model of it. All you need is a working model that affects the facets of the world you need to deal with. I still remember in the 90s…

Now imagine if grown ass adults listened to your friend assuming he understood the world fully

Re: Rodney Brooks on GPT-4

#56

Post by zone411 »

Rodney Brooks' deep learning predictions have not aged well. For example, in his 2018 blog post (http://rodneybrooks.com/forai-steps-toward-super-intelligenc...), he rates various approaches from 1-3 (with 3 being the best). Neural nets scored:

Composition: 1

Grounding: 3

Spatial: 1

Sentience: 1

Ambiguity: 2

At that time, the potential of neural nets was already very clear.

He also predicted that by 2020 we'll have popular press stories that the era of Deep Learning is over and that by 2021 VCs will figure out that for an investment to pay off there needs to be something more than "X + Deep Learning".

Re: Rodney Brooks on GPT-4

#57

Post by tail_exchange »

runthemodel wrote: post #32

At least run the model on the examples considered... Here's GPT3.5 > Can you eat a Frisbee? No, you cannot eat a Frisbee. A Frisbee is typically made of plastic, often polypropylene or similar materials, which are not meant for consumption. These materials are not digestible and can pose a choking hazard or harm your digestive system if ingested. It's important to only consume food and items that are safe and intende…

Funny. I asked GPT-4 what frisbees are most commonly made of, and it sayd polyethylene. I then asked if it was sure, and it corrected itself and said it was polypropylene. I asked again if it was sure, and it corrected itself saying it was low-density polyethylene. I kept this going for a while and it kept changing its answer.

This does seem to agree with the author. The first answer was very convincing, but not what it should have been.

Re: Rodney Brooks on GPT-4

#58

Post by jstx1 »

strofcon wrote: post #31
jstx1 wrote: post #4

> It gives an answer with complete confidence, and I sort of believe it. And half the time, it’s completely wrong. Nowhere near half in my experience. This is why we have benchmarks and metrics - so we don't need to rely on the author's opinion or on mine. > I think it’s going to be another thing that’s useful. Good.

I think you make a good point, benchmarks and metrics are indeed a better proxy for performance. Seems worth pointing out that, while "nowhere near half in [your] experience" are completely wrong, I don't take your word for it either. :-) The trouble in my view is that the only way to know that the answers you're getting are accurate and not misleading is to study up on the answers elsewhere - which is a great habit…

> If I can't know how the answer was built, or how good that answer is, there's no point asking it

In many cases, like programming for example, you can know how good the answer is - either by reading it (verifying an idea is different from coming up with it) or by testing/running code.

How the answer was built seems completely irrelevant to me, I don’t get how a useful answer produced by method x is different from a useful answer produced by method y.

Re: Rodney Brooks on GPT-4

#59

Post by famouswaffles »

zht wrote: post #55
ramraj07 wrote: post #47

Earlier quoted context omitted.

The issue with all these experts is they still think it's human nature to be able to fully understand the world before they speak about it. On the contrary it's human nature (and all animal nature) to figure out how to navigate the world without fully understanding or having a complete model of it. All you need is a working model that affects the facets of the world you need to deal with. I still remember in the 90s…

Now imagine if grown ass adults listened to your friend assuming he understood the world fully

Nobody understands the world fully so that's cool. We don't even get sense data that isn't distorted by the brain.

Re: Rodney Brooks on GPT-4

#60

Post by fcanesin »

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…

We known mathematics is the science of reality patterns, maybe if we can keep LLMs current ability in natural language and keep improving its (current poor) math abilities it will get there?...
Post reply on HN