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

I disagree with Geoff Hinton regarding "glorified autocomplete"

statmodeling.stat.columbia.edu

Post reply on HN
Print view View original

211–220 of 279 posts

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#211

Post by bigyikes »

juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

What are you talking about? This is trivially shown to be incorrect.

I just asked ChatGPT the same thing three times in a row, and it gave me three different answers, with the latter two answers being shorter and rephrased.

>I would imagine any sentient object would give a different answer every time. The first time it would give you an honest answer based on what it knows about the topic. The second time it would be a little embarrassed that you repeat the question, as if you hadn't heard the first answer. The third time it would be pissed off and think you are a troll.

Are you suggesting that a language model can't be sentient because it doesn't get annoyed like a human? That's silly.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#212

Post by nearbuy »

juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

An LLM absolutely doesn't respond the same way each time if asked the same question three times in a row, with temperature (randomness) set to zero. It responds the same way only if you start a new chat, which is a clean instance with no memory of the previous conversation. For a human, this is like if you went back in time to just before you asked the question, and asked them the same question again, in which case the person would give the same answer.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#213

Post by juhanima »

bigyikes wrote: post #211
juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

What are you talking about? This is trivially shown to be incorrect. I just asked ChatGPT the same thing three times in a row, and it gave me three different answers, with the latter two answers being shorter and rephrased. >I would imagine any sentient object would give a different answer every time. The first time it would give you an honest answer based on what it knows about the topic. The second time it would be…

ChatGPT works by cumulating the prompt. You didn't ask the same question three times. In stead you asked question q, then qq and finally qqq. Those are three different questions, which explains why you got different answers.

I'm not sure if ChatGPT also cumulates its previous answers in the context. It might do that as well. In that case the prompts would be q, qaq and qaqaq where 'q' is your question and 'a' the earlier reaction from the LLM.

The illusion of sentience comes from this. The new answers reflected what you said because the prompt was different and included the previous discussion.

This is a feature of the user interface, not the language model. The only reason why the language model would respond differently to the same input is the artificial randomness mixed with the input. Without it it would be totally deterministic and not appear sentient at all. It would still be as knowledgeable as before. Like a parrot trained to be very good at combining key words to key responses.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#214

Post by user_named »

xcv123 wrote: post #205
user_named wrote: post #191

Earlier quoted context omitted.

It is a very basic fact that LLMs have no concept of true or false, it only has an ability to look up what text data it has seen before. If you do not understand this you are in no position to discuss LLMs.

One of the most ridiculous comments I have read about LLMs here. The ~100 layer deep neural networks infer many levels of features over the text, including the concept of true and false. That is trivial for an LLM. Are you completely unaware these are based on deep neural networks? Convolutional Neural Networks don't operate by "look up" of text data.

Okay, so then tell me how does it decide whether it is true or false that Biden is the POTUS?

It's response is not based on facts about the world as it exists, but on the text data it has been trained on. As such, it is not able to determine true or false even if the response in the above example would be correct.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#215

Post by abeyer »

nearbuy wrote: post #212
juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

An LLM absolutely doesn't respond the same way each time if asked the same question three times in a row, with temperature (randomness) set to zero. It responds the same way only if you start a new chat, which is a clean instance with no memory of the previous conversation. For a human, this is like if you went back in time to just before you asked the question, and asked them the same question again, in which case t…

> For a human, this is like if you went back in time to just before you asked the question, and asked them the same question again, in which case the person would give the same answer

Is it? Would they?

You seem to assert that there's no "temperature" in human behavior... which is a reasonable theory, but not one that's universally accepted nor likely to be provable.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#216

Post by vineyardmike »

bigyikes wrote: post #211
juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

What are you talking about? This is trivially shown to be incorrect. I just asked ChatGPT the same thing three times in a row, and it gave me three different answers, with the latter two answers being shorter and rephrased. >I would imagine any sentient object would give a different answer every time. The first time it would give you an honest answer based on what it knows about the topic. The second time it would be…

> What are you talking about? This is trivially shown to be incorrect. I just asked ChatGPT the same thing three times in a row, and it gave me three different answers

Just to add color to this situation, ChatGPT has randomness built in so it generates varied answers. If you injected the same random seed each time (afaik you can’t with the gui) then you’d theoretically get the same outcome.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#217

Post by juhanima »

saurik wrote: post #209
juhanima wrote: post #204

The only reason why output from a generative LLM appears intelligent or sentient is that it parrots a random sampling of texts written by intelligent and sentient people. In order to play the game of go effectively one needs to have a model or theory of how the game of go works. That's a very simple model that can be defined by a simple formula. That's why it is fairly easy for a neural network to learn how to play t…

I don't think the question of whether an LLM that keeps getting restarted and seems to not remember things is conscious due to that lack is fair, as it feels more like suddenly making three duplicate copies of me or actively attempting to delete my memory of something... which, btw, I might not have stored in the first place: if someone has interograde amnesia, are they inherently not sentient? Even Sydney (the name…

I would argue that total anterograde amnesia would be a serious challenge for sentience, yes.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#218

Post by juhanima »

abeyer wrote: post #215
nearbuy wrote: post #212

Earlier quoted context omitted.

An LLM absolutely doesn't respond the same way each time if asked the same question three times in a row, with temperature (randomness) set to zero. It responds the same way only if you start a new chat, which is a clean instance with no memory of the previous conversation. For a human, this is like if you went back in time to just before you asked the question, and asked them the same question again, in which case t…

> For a human, this is like if you went back in time to just before you asked the question, and asked them the same question again, in which case the person would give the same answer Is it? Would they? You seem to assert that there's no "temperature" in human behavior... which is a reasonable theory, but not one that's universally accepted nor likely to be provable.

And of course "temperature" is just an euphemism for the artificial randomness that is mixed in to make the output appear more magical.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#219

Post by jeffparsons »

user_named wrote: post #191
famouswaffles wrote: post #133

Earlier quoted context omitted.

>I know my world model is fundamentally incomplete. Even more foundationally, I know that there is a world, and when my world model and the world disagree, the world wins. Yeah this isn't really true. There's not how humans work. For a variety of reasons, Plenty stick with their incorrect model despite the world indicating otherwise. In fact, this seems to be normal enough human behaviour. Everyone does it, for somet…

It is a very basic fact that LLMs have no concept of true or false, it only has an ability to look up what text data it has seen before. If you do not understand this you are in no position to discuss LLMs.

They have no inherent concept of true or false, sure. But what are you comparing them to? It would be bold to propose that humans have some inherent concept of true or false in a way that LLMs do not; for both humans and LLMs it seems to be emergent.

Re: I disagree with Geoff Hinton regarding "glorified autocomplete"

#220

Post by jeffparsons »

user_named wrote: post #214
xcv123 wrote: post #205

Earlier quoted context omitted.

One of the most ridiculous comments I have read about LLMs here. The ~100 layer deep neural networks infer many levels of features over the text, including the concept of true and false. That is trivial for an LLM. Are you completely unaware these are based on deep neural networks? Convolutional Neural Networks don't operate by "look up" of text data.

Okay, so then tell me how does it decide whether it is true or false that Biden is the POTUS? It's response is not based on facts about the world as it exists, but on the text data it has been trained on. As such, it is not able to determine true or false even if the response in the above example would be correct.

Serious question, in pursuit of understanding where you're coming from: in what way do you think that your own reckoning is fundamentally different to or more "real" than what you're describing above?

I know I don't experience the world as it is, but rather through a whole bunch of different signals I get that give me some hints about what the real world might be. For example, text.

Post reply on HN