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

And yet It Understands

borretti.me

Post reply on HN
Print view View original

221–230 of 231 posts

Re: And yet It Understands

#221

Post by YeGoblynQueenne »

marcosdumay wrote: post #220
YeGoblynQueenne wrote: post #214

Earlier quoted context omitted.

Didn't you just shift the goal posts? You care about how the structure is arrived at? Anyway the systems I have in mind do learn their own structure, I know at least three that do that. But you're not going to get machine learning without predetermined structure. In Transformers, as in all neural nets, the structure is right there in the architecture of the neural net. So what kind of "learning" without "structure" e…

> Didn't you just shift the goal posts? Nope, you changed the context from a zero-knowledge AI optimized for text into things that come with knowledge designed in. There is understanding embedded on the things with knowledge designed in, but the understanding is not done by the AI. Anyway, I'm surely curious about the systems that learn their own structure. You mean they learn the syntax rules? I'd say they have some…

I searched the thread for "zero-knowledge" and I can't find anyone having said anything like that before your above comment, and I can also see you jumping from "structure" to "knowledge" in the span of two comments, so I really think you are shifting the goal posts. More to the point, I don't think there were any goal posts at any point and you are simply prepared to make up new conditions for as long as it suits you.

So, my apologies, but I must decline to share any pointers to systems or techniques etc. You seem to have made up your mind before you even see what I'm talking about, anyway.

Re: And yet It Understands

#222

Post by fwlr »

YeGoblynQueenne wrote: post #215
fwlr wrote: post #192

Earlier quoted context omitted.

Comparing this to atheism is very illuminating, actually. In some ways this topic is very different to atheism/theism. In the ‘belief in god’ debate both sides are trying to take away the other side’s belief; it’s important to atheists to stop people believing in God, while to believers it’s important to stop people disbelieving in God, so there’s a lot of convincing being attempted, people seeking each other out to…

Just to clarify, I am an atheist and I don't mind taht people believe in god. Sorry, _God_. I respect the decision to believe and to worship. It's a fundamental element of humanity. I'm not against it at all. I just don't think that those beliefs are real, that's all. The last thing I would ever try to do is to convince someone that God doesn't exist. Why would I ever do that? As to LLMs, they are not supernatural be…

I’m not an AI researcher so my answer won’t be satisfying, but I’m pretty confident that meaning is stored in the embeddings, and gun to my head I would say understanding is hiding in the weights. My handwavey argument is that the map of activation weights between layers of neurons can store understanding, in the same way that a complete log of every transaction can store bank account balances (even though there’s no actual “this amount is in this account at this time” information anywhere in that log).

Re: And yet It Understands

#223

Post by peterfication »

fasterik wrote: post #199
MrScruff wrote: post #37

This article lines up well with my feelings on the matter. In general, people seem to understate the emergent behaviours of ML models, while overstating the uniqueness of human intelligence. I think a lot of this is down to the fact that although both systems exhibit a form of intelligence, they’re very different. LLMs deliver mastery of natural language that would normally be a signal for a highly intelligent human.…

I agree that we don't know what is possible in hypothetical future models, but claims about understanding in the current models are overblown (at least by people like the article author). For example, I gave GPT-4 some chess games and asked it to evaluate the final move of each game as good or bad and explain its evaluation. It clearly had some understanding of the board state and what makes a move good or bad, but t…

This is probably something that will be fixed with an OpenAI to "some chess AI"-plugin. Then for the current model it's just enough to understand that it's a chess problem.

Re: And yet It Understands

#224

Post by YeGoblynQueenne »

fwlr wrote: post #222
YeGoblynQueenne wrote: post #215

Earlier quoted context omitted.

Just to clarify, I am an atheist and I don't mind taht people believe in god. Sorry, _God_. I respect the decision to believe and to worship. It's a fundamental element of humanity. I'm not against it at all. I just don't think that those beliefs are real, that's all. The last thing I would ever try to do is to convince someone that God doesn't exist. Why would I ever do that? As to LLMs, they are not supernatural be…

I’m not an AI researcher so my answer won’t be satisfying, but I’m pretty confident that meaning is stored in the embeddings, and gun to my head I would say understanding is hiding in the weights. My handwavey argument is that the map of activation weights between layers of neurons can store understanding, in the same way that a complete log of every transaction can store bank account balances (even though there’s no…

Well, the idea with embeddings is, indeed, that they are a representation of meaning, specifically a spatial representation of the meaning of words. The theory is that if two words are close together in "embedding space", their meaning is also close together.

The problem with that has always been that embeddings can be said to encode relative meaning, but not absolute meaning. That is, to understand the meaning of a word A with reference to a word B, you need to know the meaning of B. And the only entities that can do that -tell you what a word means - are humans. So a neural net may learn all the embeddings it likes, it's missing a foundation of meaning on which to base them.

To give an analogy, imagine trying to understand what a word means, in a language you don't know, given only a dictionary of that language. Every word in the dictionary is explained, except it's explained in a language you don't know. You can see that some words are explained in terms of other words, but that doesn't really help much because you don't know the meaning of the other words in the first place. It's lemmas all the way down.

Or suppose I gave you an encrypted string. You can learn the string by heart and repeat it as many times you like, but you won't know what it says until you can decrypt it. So maybe embeddings encode meaning, but where is it that they decode it?

The same goes with neural net weights. The problem with all those techniques -and, to be fair, with every other approach to AI so far- is that we're missing a foundational encoding of meaning that we can relate other representations to. Some people think we can get over this by training neural nets on real-world data, or make robots and let them experiment with the world etc, but I don't know that anyone has really shown that this works yet.

So, maybe neural nets can store understanding in their weights, but we don't know how to put it there, yet.

Re: And yet It Understands

#225

Post by zshrdlu »

jmoak3 wrote: post #219

The more AI develops the less omnipotent I feel about human level intelligence. Not once had I ever considered anything could exist as intelligent as a person. I’m not saying GPT4 is there, but to say nothing equal or greater than us will ever exist anywhere in the universe? I wouldn’t take that bet nowadays. I’ve cut meat from my diet over these thoughts, it makes me want to be a slightly better steward of the other…

> I’ve cut meat from my diet over these thoughts, it makes me want to be a slightly better steward of the other intelligences we’ve conquered.

I'm curious if you've made any changes to your life(style) to reduce the suffering of fellow human beings.

Re: And yet It Understands

#226

Post by jmoak3 »

zshrdlu wrote: post #225
jmoak3 wrote: post #219

The more AI develops the less omnipotent I feel about human level intelligence. Not once had I ever considered anything could exist as intelligent as a person. I’m not saying GPT4 is there, but to say nothing equal or greater than us will ever exist anywhere in the universe? I wouldn’t take that bet nowadays. I’ve cut meat from my diet over these thoughts, it makes me want to be a slightly better steward of the other…

> I’ve cut meat from my diet over these thoughts, it makes me want to be a slightly better steward of the other intelligences we’ve conquered. I'm curious if you've made any changes to your life(style) to reduce the suffering of fellow human beings.

I cut out eating people years ago

Re: And yet It Understands

#227

Post by kthejoker2 »

detrites wrote: post #211
rthrfrd wrote: post #153

Earlier quoted context omitted.

I’m curious as to what “higher order emergent behaviours” you believe you are observing? Thanks.

Imagine ChatGPT 5 or 6, where it's at the point that it's entirely indistinguishable from a human, by any one of us. At this point AI can mimmic every aspect of being a human: emotions, reasoning, intelligence, memory. Will you, at that point - which isn't far away - still claim that because it's based on matrix multiplication it can't possibly be doing any of those things and none of the behaviours fooling everyone…

Sorry for the late reply, I think you are arguing two points, one I agree with and one I do not.

I agree it has abilities, we can even call them behaviors.

But ... yeah, every single one of its responses is just the output of a bunch of matrix multiplciation. I can even do the math on paper (it just would take a while ...)

we anthromorphize many human-like behaviors in non-humans, AI is by its nature designed to be more human like than other mon-humans, of course its responses which look like human behaviors (mimicking implies agency) will "fool" people into anthropomorphizing it as well.

Re: And yet It Understands

#228

Post by zshrdlu »

jmoak3 wrote: post #226
zshrdlu wrote: post #225

Earlier quoted context omitted.

> I’ve cut meat from my diet over these thoughts, it makes me want to be a slightly better steward of the other intelligences we’ve conquered. I'm curious if you've made any changes to your life(style) to reduce the suffering of fellow human beings.

I cut out eating people years ago

That's not good for the environment!

Re: And yet It Understands

#229

Post by zshrdlu »

Nevermark wrote: post #187

Two-layer neural networks are universal approximators. Given enough units/parameters in the first layer, enough data, and enough computation, they can model any relationship. (Any relationship with a finite number of discontinuities. Which covers everything we care about here.) But more layers, and recurrent layers, let deep learning models learn complex relationships with far fewer parameters, far less data and far…

> So pointing out that an LLM has trouble with some concept doesn’t mean anything.

Why? We do the same with children, animals, and people (with severe head trauma for example). Why should AI get special treatment?

We're happy to test if crows and dolphins can do arithmetic and just all sorts of cognitive hoops.

Re: And yet It Understands

#230

Post by smitty1e »

doctor_eval wrote: post #195
smitty1e wrote: post #98

Earlier quoted context omitted.

Well, if we're to believe that thr Philosopher's Stone has arrived, why should it not have oracular power? I am regularly assured by the Climatistas that the apocalypse is at hand. What saith ChatGPT?

Literally nobody is saying that about ChatGPT. Not sure what the climate has to do with this, other than maybe signalling whose team you're on.

Well, the thrust of the headlines seems to be that ChatGPT is some sort of Delphic Oracle.

Therefore, let it prophesy.

Post reply on HN