Theory of Mind May Have Spontaneously Emerged in Large Language Models • Hacker News Board

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#51

Post by didntreadarticl » Thu, Feb 09, 2023, 10:09 PM UTC

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

That is just not in evidence.

Seems like its nailing it to me. You ask about a scenario and it gives an appropriate answer.

We have evidence that LLMs build models of the things they are learning about. Have a look at this paper:

Do Large Language Models learn world models or just surface statistics?

https://thegradient.pub/othello/

previously discussed https://news.ycombinator.com/item?id=34474043

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#52

Post by pdonis » Thu, Feb 09, 2023, 10:11 PM UTC

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

> In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person.

I wish I could upvote this 1000 times. It is the core issue that all the hype surrounding LLMs consistently fails to address or even acknowledge.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#53

Post by canniballectern » Thu, Feb 09, 2023, 10:11 PM UTC

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

Of course, today's LLMs only appear to have theory of mind at first glance and fall apart under closer scrutiny. But if they can continue to become more and more accurate replicas of the real thing, I don't think it matters at all.

There's no way to know for sure that anyone other than yourself experiences consciousness. All you can do is judge for yourself that what they're describing matches closely enough with your own experiences that they're probably experiencing the same thing you are.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#54

Post by visarga » Thu, Feb 09, 2023, 10:11 PM UTC

mlajtos wrote: post #12

This is intriguing. Could it be simply explained by introducing ToM (or ToM-like) training data? Since all DaVinci models are 175B parameters, the extra training or training data must be the reason for the improvement. Do we know how different DaVinci models are trained?

> the extra training or training data must be the reason for the improvement

People are blinded by the model size and often forget about the data. I think somehow intelligence is encoded in language, including theory of mind.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#55

Post by kenjackson » Thu, Feb 09, 2023, 10:11 PM UTC

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

It seems obvious that LLMs don’t understand a bag the way we do. It’s never seen a bag. Or held one. But if you equip it with the same IO as humans, how different would it be? Probably still pretty different, but light years closer than what we had 20 years ago.

Also humans are good at next word guessing their own next word. Each person has been trained on a different set of data, so it’s no surprise that they wouldn’t be able to guess other people’s next words.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#56

Post by space_fountain » Thu, Feb 09, 2023, 10:11 PM UTC

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

I wonder every time I see this take what it would mean under this definition of knowing things for a machine learning algorithm to ever know something. I find that especially important because to every appearance we are a machine learning algorithm. I don’t know how different the sort of knowing this algorithm has to the sort of knowing a human has, but you’re far more confident than I am that it’s a difference of kind rather than degree.

Some interesting facts that point to it being a difference of degree. LLM are actually are more accurate when asked to explain their thinking. They make similar mistakes to humans intuitive reasoning.

It might help to define what we even mean by knowing things. To me being able to make novel predictions that require the knowledge is the only definition one could use that doesn’t run into the possibility of deciding humans don’t actually know anything

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#57

Post by pdonis » Thu, Feb 09, 2023, 10:13 PM UTC

spuz wrote: post #48

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

> So the tasks remain useful and accurate in testing ToM in people because people can't perform statistical regressions over billion-token sets and therefore must generate their thoughts the old fashioned way. Is it not also possible that the study suggests that the human mind actually operates as a statistical regression over billions of data points rather than through some kind of Baysian logic? You say humans are…

> Is it not also possible that the study suggests that the human mind actually operates as a statistical regression over billions of data points rather than through some kind of Baysian logic?

No. Human minds have semantic relationships to the rest of the world that LLMs do not have. The comparisons being made between the two, not just in this paper but in all of the hype surrounding LLMs, are simply invalid. But they sure help in collecting more funding.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#58

Post by robertlagrant » Thu, Feb 09, 2023, 10:13 PM UTC

space_fountain wrote: post #56

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

I wonder every time I see this take what it would mean under this definition of knowing things for a machine learning algorithm to ever know something. I find that especially important because to every appearance we are a machine learning algorithm. I don’t know how different the sort of knowing this algorithm has to the sort of knowing a human has, but you’re far more confident than I am that it’s a difference of ki…

> It might help to define what we even mean by knowing things

This is it, I think. It's interesting that we now have a practical example to point at when asking formerly-abstruse philosophical questions.

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#59

Post by robertlagrant » Thu, Feb 09, 2023, 10:16 PM UTC

canniballectern wrote: post #53

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

Of course, today's LLMs only appear to have theory of mind at first glance and fall apart under closer scrutiny. But if they can continue to become more and more accurate replicas of the real thing, I don't think it matters at all. There's no way to know for sure that anyone other than yourself experiences consciousness. All you can do is judge for yourself that what they're describing matches closely enough with you…

> But if they can continue to become more and more accurate replicas of the real thing, I don't think it matters at all.

So, I suppose I'd ask: what does "matter" mean here? If you knew that everyone you loved had been destroyed and been replaced by exact replicas, would that matter?

Re: Theory of Mind May Have Spontaneously Emerged in Large Language Models

#60

Post by pdonis » Thu, Feb 09, 2023, 10:16 PM UTC

canniballectern wrote: post #53

lsy wrote: post #33

This highlights one of the types of muddled thinking around LLMs. These tasks are used to test theory of mind because for people, language is a reliable representation of what type of thoughts are going on in the person's mind. In the case of an LLM the language generated doesn't have the same relationship to reality as it does for a person. What is being demonstrated in the article is that given billions of tokens o…

Of course, today's LLMs only appear to have theory of mind at first glance and fall apart under closer scrutiny. But if they can continue to become more and more accurate replicas of the real thing, I don't think it matters at all. There's no way to know for sure that anyone other than yourself experiences consciousness. All you can do is judge for yourself that what they're describing matches closely enough with you…

> There's no way to know for sure that anyone other than yourself experiences consciousness. All you can do is judge for yourself that what they're describing matches closely enough with your own experiences that they're probably experiencing the same thing you are.

That judgment is not just based on the words other people use. It is based on knowing that other people's brains and minds have the same sort of semantic relationships to the rest of the world that yours do. And those relationships can be tested by checking to see if, for example, the other person uses the same words to refer to particular objects in the real world that you do, or if they react to particular real-world events in the same way that you do.

You can't even test any of this with an LLM because the LLM simply does not have the same kind of semantic relationships with the rest of the world that you do. It has no such relationships at all.