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

211–220 of 231 posts

Re: And yet It Understands

#211

Post by detrites »

rthrfrd wrote: post #153
detrites wrote: post #140

Earlier quoted context omitted.

How about those of us with an intimate understanding of how LLM's are trained and exactly how transformers etc function who accept that the resultant models higher order emergent behaviours , are exactly that and not merely "coincident"?

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 are actually there?

How something is built does not limit its abilities. Only what something can do limits its abilities.

AI has certain abilities now.

Whether they emerged, were built-in by careful design, or don't even exist and just "seem" to, ultimately isn't even relevant to the most important fact of the matter, which is simply that it has those abilities.

To answer your question more directly, the emergent behaviours I'm seeing are: all of them.

That is, anything it seems to be able to do well enough to be useful and reliable, has emerged.

Re: And yet It Understands

#212

Post by _xnmw »

marcosdumay wrote: post #133
_xnmw wrote: post #127

I still hold that it doesn't "understand". Even if it answered all questions perfectly, stopped making mistakes, and produced fully working programs better than the best crack developer teams, that still doesn't mean it "understands". "Understanding" is not an output, it's a process, that is sometimes (but not always) measured by its output.

If it stopped making mistakes and produced complete fully working programs, no, there would be no way to say it doesn't understand. Yes, "understanding" is a process, but it's not well defined. And anyway, if it's a requirement for those things, and the AI did those things, then the only possibility is that the AI has this process in some way. But well, our current AIs do not produce complete programs, nor fully work…

Quality of output does not mean that the process is genuine. A well-produced movie with good actors may depict a war better than footage of an actual war, but that is not evidence of an actual war happening. Statistical LLMs are trying really hard at "acting" to produce output that looks like there is genuine understanding, but there is no understanding going on, regardless of how good the output looks.

Re: And yet It Understands

#213

Post by _xnmw »

pornel wrote: post #182
_xnmw wrote: post #127

I still hold that it doesn't "understand". Even if it answered all questions perfectly, stopped making mistakes, and produced fully working programs better than the best crack developer teams, that still doesn't mean it "understands". "Understanding" is not an output, it's a process, that is sometimes (but not always) measured by its output.

To me setting the bar so high makes that definition of "understanding" meaningless. If AI's outputs were literally indistinguishable from something that truly understands these concepts, then there's no practical difference if it really "understands". This reduces the concept of understanding to some abstract inconsequential property.

> If AI's outputs were literally indistinguishable from something that truly understands these concepts, then there's no practical difference if it really "understands". This reduces the concept of understanding to some abstract inconsequential property.

Yes, there is a practical difference. It has consequences in terms of legality, blame, and justice, the other does not. If an AI really "understands" things, then it is responsible for its actions and must be taken to court for crimes, and can be sued for mistakes in the output that result in losses (like any lawyer or professional).

Re: And yet It Understands

#214

Post by YeGoblynQueenne »

marcosdumay wrote: post #202
YeGoblynQueenne wrote: post #164

Earlier quoted context omitted.

>> If it stopped making mistakes and produced complete fully working programs, no, there would be no way to say it doesn't understand. There exist program synthesis systems that always produce "complete fully working programs" and that in fact cannot make any mistakes because they are based on algorithms with strong theoretical guarantees (like mathematical proofs) of their correctness. You give them a set of example…

> a program synthesis system that never makes mistakes "understands" Yes, I'm saying if it satisfies random requirements and synthesizes full working programs, then it has some form of understanding in it. If creates that understanding through learning, it's very possibly a general form of AI (I'm not certain of this part). I've never seen any one that arrived at structure by learning. It's always programed in, what…

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" exactly are you looking for?

Re: And yet It Understands

#215

Post by YeGoblynQueenne »

fwlr wrote: post #192
YeGoblynQueenne wrote: post #156

Earlier quoted context omitted.

>> Those who deny that ChatGPT understands have to move their goalposts every few weeks; OpenAI’s release schedule seems to be slightly faster, so in time it seems even the fastest-moving goalposts will be outrun by the LLMs. No, they don't. It doesn't matter how many releases OpenAI makes, there is still no alternative explanation needed for the behaviour of their systems than the fact that their systems are languag…

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 beings that are the subject of belief, or even disbelief. They are technological artifacts created by humans, machines that perform a task they were created to perform. We created the machines, we know how they are made, and we know what they can do. For example, if you were to argue that a car can understand transport, or an airplane, flight, just because they are very complex technological artifacts and they behave in ways that are not easy to predict, or even explain, then I'd have to say, similar to what I say about LLMs: "an airplane is a machine designed to fly, not understand flight, so if you think it understands flight it's up to you to show me how its mechanisms allow it to do that".

In particular I would ask you this because I am just now polishing up a PhD in AI (although not the kind you read in the news) so this is my field of expertise and I believe it is reasonable for me to have questions, when someone makes assertions about the functioning of systems that I have studied.

In short, it's my job to know and to understand how those systems work, and if you say they 're working somehow otherwise, then of course I'd like to know why you say that. So, yes, I'm going to want to know where "understanding" has been hiding from me all this time, and I'd like very precise instructions to find it myself, please, otherwise I'm going to assume you're making it up as you go along.

No offense meant. I have to do that as part of my job (i.e. reviews). Perhaps it's a "deformation professionelle", eh?

Re: And yet It Understands

#216

Post by YeGoblynQueenne »

rstuart4133 wrote: post #200
YeGoblynQueenne wrote: post #156

Earlier quoted context omitted.

>> Those who deny that ChatGPT understands have to move their goalposts every few weeks; OpenAI’s release schedule seems to be slightly faster, so in time it seems even the fastest-moving goalposts will be outrun by the LLMs. No, they don't. It doesn't matter how many releases OpenAI makes, there is still no alternative explanation needed for the behaviour of their systems than the fact that their systems are languag…

I agree there is a connection between ChatGPT, god and atheists but I don't think it's the one you point out. The commonality "general intelligence" and "god" share is there is no agreed definition of either. Science has forced a steady retreat in the number of things that can be attributed to god. But that shouldn't worry theists, because Godel's incomplete theorem means there are some things we will can never under…

I don't know if you lost any debates there. If you were having those debates for a long time, you may have heard of Russel's Teapot:

Russell's teapot is an analogy, formulated by the philosopher Bertrand Russell (1872–1970), to illustrate that the philosophic burden of proof lies upon a person making empirically unfalsifiable claims, rather than shifting the burden of disproof to others.

https://en.wikipedia.org/wiki/Russell%27s_teapot

So there's nothing there to disprove. Anyone who wants to convince you of the existence of the Gods they believe in, that person has the responsibility to convince you. Not you, them.

But I think it's a bad idea to try to convince people to believe, or disbelieve things anyway. Why do that? Faith is a great thing to have, life is hard and there's horrible things that happen during it. There's nothing more powerful than knowing in your heart of hearts that there's someone big, and almighty, watching over you.

Leave people believe. You're not doing them a favour by trying to refute their faith.

Re: And yet It Understands

#217

Post by dallyo »

kthejoker2 wrote: post #118
trasher9009 wrote: post #5

Is this real? Haha I just may not understand at this level.. lol. User: are green potatoes poisonous? Sydney: Green potatoes can be toxic to humans [1,2]. The green color may indicate the presence of a toxin called solanine[1,2]. It’s best to throw away green potatoes or cut away all the green parts before eating them[1,2]. Why do you ask? Are you planning to cook or eat green potatoes? User: my toddler ate green pot…

Assume it is real: it is regurgitating tokens based on what the collective corpus of text it was trained on would most likely reply to a similar scenario. I wouldn't be surprised if similar wording is not in the call scripts of poison control hotlines. Not sure why this particular example is bring held up as some form of "understanding." People's inability (or unwillingness) to understand how LLMs are trained and how…

"Please don't give up on your child" in hotline scripts? I think that's highly unlikely.

Re: And yet It Understands

#218

Post by pornel »

_xnmw wrote: post #213
pornel wrote: post #182

Earlier quoted context omitted.

To me setting the bar so high makes that definition of "understanding" meaningless. If AI's outputs were literally indistinguishable from something that truly understands these concepts, then there's no practical difference if it really "understands". This reduces the concept of understanding to some abstract inconsequential property.

> If AI's outputs were literally indistinguishable from something that truly understands these concepts, then there's no practical difference if it really "understands". This reduces the concept of understanding to some abstract inconsequential property. Yes, there is a practical difference. It has consequences in terms of legality, blame, and justice, the other does not. If an AI really "understands" things, then it…

That is a huge leap into personhood and very human concepts that depend on having volition, motives, feelings, and self-awareness or consciousness. I don't think these are necessary connected to understanding. IMHO it's possible to have full understanding of all of the concepts in the world while only seeing the world as a data snapshot, without being an entity capable of participating in it, without having any feelings, self-awareness, or even any sensory inputs or memory of own experiences.

Laws, courts, and their punishments are for humans, and they're likely to be absurdly bad at dealing with a personified AI. AI won't feel fear or remorse, may not have self-preservation as a goal, so it may not care about any threats at all. It may not even have sensory input needed to be aware of being punished.

Re: And yet It Understands

#219

Post by jmoak3 »

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 intelligences we’ve conquered.

It feels like I’ve just realized the earth orbits the sun and not the other way around, so to speak.

Re: And yet It Understands

#220

Post by marcosdumay »

YeGoblynQueenne wrote: post #214
marcosdumay wrote: post #202

Earlier quoted context omitted.

> a program synthesis system that never makes mistakes "understands" Yes, I'm saying if it satisfies random requirements and synthesizes full working programs, then it has some form of understanding in it. If creates that understanding through learning, it's very possibly a general form of AI (I'm not certain of this part). I've never seen any one that arrived at structure by learning. It's always programed in, what…

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 understanding of the language syntax, but well, unless they can take a requirement and produce working code, they don't understand programing. (And well, if the requirement is in a formal language, they only understand formal transformations, that is, your AI understands compiling. What honestly, is quite limited, but would still be impressive.)

Post reply on HN