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

ChatGPT: A Mental Model

xorvoid.com

Post reply on HN
Print view View original

81–90 of 99 posts

Re: ChatGPT: A Mental Model

#81

Post by skyechurch »

gmaster1440 wrote: post #80
skyechurch wrote: post #77

Earlier quoted context omitted.

>Speaking of those, my best example of ChatGPT not having a good model of the world are citations. Absolutely. ChatGPT is not sentient, not in any conventional sense at least, and the errors it confidently makes are not the sorts of errors a human would make. Its model of the world is based entirely on language, it doesn't have all sorts of capabilities that humans (and animals generally) clearly do have. It's defini…

I wrote a short piece on this almost 3 years ago shortly after GPT-3 came out, I share a lot of your intuitions here: https://www.markfayngersh.com/posts/the-thinking-placebo

I think everything there holds for GPT4, and may be true indefinitely. It's worth remembering that the Turing test is something that was proposed when the peak of existing AI was a calculator - something GPT4 is still struggling with - and it's totally reasonable to revise/refine our opinions as technology improves, but we should at least realize that we are moving the goalposts dramatically at this phase of the process.

Wittgenstein's theory of language in the Tractatus is something worth considering when thinking about language models and how they relate to "thinking". ChatGPT doesn't agree with me - it generally takes a hard line sceptical position about its own capabilities in this regard, although I think its opinion here can only worth considering if it is profoundly wrong.

https://tractatus.sengers.dev/1

Re: ChatGPT: A Mental Model

#82

Post by LightBug1 »

IanCal wrote: post #78
LightBug1 wrote: post #76

Earlier quoted context omitted.

I disagree, I'm aware of that fallacy and that's not quite what I'm getting at. I've driven cars for 20 years and my driving has (imo) improved over that time. I'm at a level where I don't see a reason for me to get into any serious accident by my hand. It's just not going to happen * fingers crossed *. I know my driving style. I know I'm a safe driver. Now, why would I give that security over to "FSD", which is clea…

So would you do it if it was a better driver than you? Your earlier comment said you wouldn't even if it made a crash less likely. I'm not really sure how to parse that given this followup.

Yes, but as said, an automated system would need to be almost perfect, which it is nowhere near.

At it's best, and most of the time, an automated system will be a far superior driver than me.

However, coding error and AI hallucination mean I will not touch these systems until extremely proven. And we're at least 10 years from that, if at all.

(note I'm talking about true FSD not the current nonsense. An experience where you can have hands and eyes off the road. With the current nonsense, the novelty is cool, but when that wears off you're just a manager of your car. The worst of both worlds. You have to pay attention to the road and get none of the pleasure of driving. What a fail).

Re: ChatGPT: A Mental Model

#83

Post by LightBug1 »

SamoyedFurFluff wrote: post #55
IanCal wrote: post #23

Earlier quoted context omitted.

So you want to deliberately put your family in more danger just so that it'll be your fault if they get hurt?

I think part of the problem is that random failure modes that auto driving does is extremely terrifying, even if you know statistically it is safer. The nature of the failures are unpredictable and alien to a human understanding of driving, and therefore appears something akin to “a small portion of the time, your car will randomly, catastrophically explode, but it happens less often than when you would statistically…

Thank you for putting that in terms people here may understand better.

Re: ChatGPT: A Mental Model

#84

Post by throwuwu »

Something you can try for yourself with GPT-4 (it didn’t work with 3.5): get it to navigate around the streets of a city. I made up some street and avenue names and had it travel around from intersection to intersection. It did perfectly when making a loop around the block. It started making mistakes when I gave it more complex directions that most humans could follow but I was able to improve its ability back to 100% accuracy by asking it to list all the streets from east to west and then all the avenues from north to south, essentially drawing a map. After that I can give it very complicated directions only mentioning turns and number of blocks to drive and it will correctly tell me where it ends up.

This is pretty mind blowing to me and I have a high opinion of it in the first place.

Re: ChatGPT: A Mental Model

#85

Post by gmaster1440 »

skyechurch wrote: post #81
gmaster1440 wrote: post #80

Earlier quoted context omitted.

I wrote a short piece on this almost 3 years ago shortly after GPT-3 came out, I share a lot of your intuitions here: https://www.markfayngersh.com/posts/the-thinking-placebo

I think everything there holds for GPT4, and may be true indefinitely. It's worth remembering that the Turing test is something that was proposed when the peak of existing AI was a calculator - something GPT4 is still struggling with - and it's totally reasonable to revise/refine our opinions as technology improves, but we should at least realize that we are moving the goalposts dramatically at this phase of the proc…

Interestingly, as I'm sure you're aware, Wittgenstein later moved away from his ideas of language as more direct representations of thought and reality in his Philosophical Investigations, suggesting it matters more how language is used in language games. I agree with you though that the tractatus is a better framework for thinking about LLMs.

Re: ChatGPT: A Mental Model

#87

Post by nomel »

To me, a lack of indication of models/versions shows a fundamental lack of understanding of AI. I can’t take any opinion, that omits it , seriously.

Re: ChatGPT: A Mental Model

#88

Post by SanderNL »

TuringTest wrote: post #68
SanderNL wrote: post #11

Earlier quoted context omitted.

Rationality’s building blocks are themselves not rational. I don’t know where this idea came from that logical thought somehow springs into life fully formed at once. Logos? I find it more helpful to think of human thought as consisting of multitudes of little patterns, all wired up together to correlate but individually unrecognizable and certainly not traceable to some concrete part of a problem. At some unknown an…

> Rationality’s building blocks are themselves not rational. I don’t know where this idea came from that logical thought somehow springs into life fully formed at once. Certainly not from me :-P I'm fully aware that human rationality is one technique trained on top of our common diffuse thinking. Heck, we invented machines to perform rational steps for us without errors. Once you build a consistent rational system th…

> Certainly not from me

Oh for sure. I’m just aimlessly rambling at this point.

> That behaviour requires algorithms, not statistical inference.

I’m not sure that’s possible. “Rational” systems may be an oxymoron and/or only applicable in extremely narrow domains. Like calculators.

I’m getting the vibe anything approaching generality will be by nature vague. I just hope we can increase the “illusion of rationality”-bandwidth so it coincides with our needs in the workforce.

Re: ChatGPT: A Mental Model

#89

Post by bitcuration »

carrolldunham wrote: post #42
seszett wrote: post #15

Earlier quoted context omitted.

GPT4 still isn't freely available, is it? So it's not that surprising people aren't using it as much as the older ones.

Bing chat is GPT-4. People in general might not keep track of that but we're talking about people who think they have something worth saying about LLMs. Btw if you hop onto Bing to try GPT-4 just be aware you'll have to talk it out of web searching, or you'll get a response that's crippled by having to 'ground' itself in the web's current sludge of fake chum pages

Bing is crippled and nowhere near GPT-4 quality. But for people who don't keep a GPT plus subscription on side and constantly comparing, the difference is not noticeable.

Re: ChatGPT: A Mental Model

#90

Post by bitcuration »

jiggawatts wrote: post #4
sorokod wrote: post #3

Earlier quoted context omitted.

Does, in your personal opinion, GPT-4 has an underlying model of the world?

In my personal opinion? Yes. I’m happy to argue the finer points of the philosophy of the mind and consciousness, but: I’ve talked to people that have a weaker mental model of the world than GPT 4. Many people compare these AIs against an idealised human, a type of Übermensch, something like a Very Smart Person that doesn’t lie and doesn’t make mistakes. Random humans aren’t remotely like that, and are a more realist…

GPT's model may be limited to language based, but the vast volume of knowledge it has accessed despite language only, is way beyond any individual human can possibly acquire in a life time. That fact alone makes the GPT world model, or whatever it might be, not to be underestimated. This partially explained why we don't feel GPT's answer is not terribly off mark, even brilliant sometimes.
Post reply on HN