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

Microsoft Says New AI Shows Signs of Human Reasoning

nytimes.com

Post reply on HN
Print view View original

31–40 of 40 posts

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#31

Post by nicd »

hulitu wrote: post #21
hersko wrote: post #10

Not surprising at all. I throw a wall of code at GPT-4, write a paragraph of what i want it to do (I couldn't spend the time to figure out a good solution), and within a minute I had a great functioning solution perfectly integrated into my existing code. The surprising thing is not that it was able to code a good solution, it's that it was able to understand the specifications of how i wanted it to work and how it s…

Does it compile ? (the code generated)

In my experience with ChatGPT-4, I'll get a full 2-300 line class, with maybe 1-2 compilation errors (often a missing import or invalid function invocation). In most cases, I can copy the compilation error back into the chat, GPT will apologize, fix it, and it'll run correctly the second time.

It's obvious that this will be faster and more effective once a GPT4-class model is tightly integrated into the run-compile-test loop.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#32

Post by ActorNightly »

creatonez wrote: post #12
ActorNightly wrote: post #7

I mean, ability to parse information, compress it into the latent space, and then regenerate it is what humans do, but I wouldn't call that reasoning.

Is reasoning about the information the best way to compress it into latent space?

No. Its about figuring out the state machine.

For example, given a set of text, an LLM will predictably pick like 1 out of 3 possibilities for the next word, and that word affects it picking the next word and so on. So its essentially turning what would otherwise be a giant look up table into a compressed markov model representation.

Reasoning is the ability to generate information that is not currently in that model. For example, if you take a neural net and train it on all the properties of fluids, it should be able to give you answers about aerodynamics if it can truly reason.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#33

Post by NumberWangMan »

ilaksh wrote: post #29

The problem is that people aren't able to use language consistently and precisely, and are conflating all kinds of human/animal characteristics together. GPT-4 really does do human-like reasoning. And it's clearly quite general purpose within its limitations. But it doesn't have many other aspects of humans/animals such as self-direction, high bandwidth senses, a stream of subjective experience, emotions, certain typ…

I mean, if it’s reasoning, we may be very close to AGI, and if we are, that’s a big problem. We’re on a trajectory for these models to become a lot smarter than us, and we’re not ready to deal with that.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#34

Post by dTal »

gumballindie wrote: post #14
Delotosex wrote: post #6

Earlier quoted context omitted.

There is not much to fall for when you can test it for free already. I also watches the writer of the papers talk and it was a good talk. I also saw the progress of ml in the last 5 years and I have never seen something progressing that fast (besides Smartphones perhaps). So it's not far fetched that got 4 is not at the ceiling of doable. The opposite: chatgpt makes it much easier to get funding, the race is on

AI is not at the stage of smartphones, it’s at the stages of the early transistor. It will grow because we have the data and the hardware to make it grow. But it appears that human intelligence is on a reverse trend, which is why it’s so easy to sell the idea that ai is intelligent and all of the bad scifi around it.

> But it appears that human intelligence is on a reverse trend

A dangerous myth. Idiocracy was an entertainment film with a side order of eugenics, not a documentary.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#35

Post by creatonez »

ActorNightly wrote: post #32
creatonez wrote: post #12

Earlier quoted context omitted.

Is reasoning about the information the best way to compress it into latent space?

No. Its about figuring out the state machine. For example, given a set of text, an LLM will predictably pick like 1 out of 3 possibilities for the next word, and that word affects it picking the next word and so on. So its essentially turning what would otherwise be a giant look up table into a compressed markov model representation. Reasoning is the ability to generate information that is not currently in that model…

I'm unconvinced. You've described that it transforms an input to an output, but you've ignored the possibility that the hidden layers in the neural network have discovered that generalizations are the best approach to predicting in entirely novel situations. Long before GPT came about, machine learning researchers have discovered (in research examining the purpose each neuron serves) emergent generalizations in image classifier models.

To use the example in the article, the answer to

"Here we have a book, nine eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner"

is not something the original 'lookup table' could possibly help you with, other than by manually reasoning about how the various parts of a 'concept web' fit together. This question represents a completely unfilled gap in the training data.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#36

Post by ActorNightly »

creatonez wrote: post #35
ActorNightly wrote: post #32

Earlier quoted context omitted.

No. Its about figuring out the state machine. For example, given a set of text, an LLM will predictably pick like 1 out of 3 possibilities for the next word, and that word affects it picking the next word and so on. So its essentially turning what would otherwise be a giant look up table into a compressed markov model representation. Reasoning is the ability to generate information that is not currently in that model…

I'm unconvinced. You've described that it transforms an input to an output, but you've ignored the possibility that the hidden layers in the neural network have discovered that generalizations are the best approach to predicting in entirely novel situations. Long before GPT came about, machine learning researchers have discovered (in research examining the purpose each neuron serves) emergent generalizations in image…

>is not something the original 'lookup table' could possibly help you with,

Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic.

Lets say I ask you the following question "Here we have a flibert, nine moltice, a grook, a seerik and a lopo. Please tell me how to stack them onto each other in a stable manner". You would be able to figure out how to stack them, given the fact that you have the ability to "reason" and figure out what those items are in the first place. In the same way, an AI that can reason would be able to do the same.

Just like transformers were revolutionary because they essentially are representative of a process inside human brains where we learn to pay attention to select features, there will be some new architecture in the next 10 years that will encode this ability to assign semantic properties to information and operate on those properties, which will then allow for things like AI to ask clarifying questions.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#37

Post by creatonez »

ActorNightly wrote: post #36
creatonez wrote: post #35

Earlier quoted context omitted.

I'm unconvinced. You've described that it transforms an input to an output, but you've ignored the possibility that the hidden layers in the neural network have discovered that generalizations are the best approach to predicting in entirely novel situations. Long before GPT came about, machine learning researchers have discovered (in research examining the purpose each neuron serves) emergent generalizations in image…

>is not something the original 'lookup table' could possibly help you with, Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic. Lets say I ask you the following question "Here we have a flibert, nine moltice, a grook, a seerik and a lopo. Please tell me how to stack them onto eac…

> Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic.

This is very reductionist. Of course everything is numbers under the hood. But it's entirely orthogonal to the question. Numbers can encode all of the complexity needed for anything that can do reasoning. Numbers in the form of a neural network can execute arbitrary algorithms, given enough layers.

> Lets say I ask you the following question "Here we have a flibert, nine moltice, a grook, a seerik and a lopo. Please tell me how to stack them onto each other in a stable manner". You would be able to figure out how to stack them, given the fact that you have the ability to "reason" and figure out what those items are in the first place. In the same way, an AI that can reason would be able to do the same.

I really don't understand the purpose of this thought experiment. A human is able to figure this out because they can see what those objects are. Is your objection to language model reasoning that they are not multi-modal?

> Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#38

Post by coolhand2120 »

wizofaus wrote: post #28
coolhand2120 wrote: post #18

If anyone has read Blindsight by Peter Watts the hero encounters an alien that is pure reflex, no cognition. The alien is compared to the Chinese room thought experiment ( https://en.wikipedia.org/wiki/Chinese_room ). ChatGTP is the Chinese room. As you peel the onion you discover that some of the arguments around thought are really just semantics. My personal experience shows me that ChatGTP 4 is something of a fool…

I thought I was pretty knowledgeable in music theory but I'd not heard of the H transformation. I'm a bit nonplussed as to why it's even a thing, and indeed Google doesn't even find any hits in the first page describing it that I can see. https://en.wikipedia.org/wiki/Neo-Riemannian_theory describes it as "the H relation", which sounds a bit more familiar, though I admit Neo- (vs regular!) Riemannian theory isn't som…

This is a wonderful deep explanation of Neo-Riemannian theory. This IS the "secret sauce" of modern cinematic music IMHO.

https://alpof.wordpress.com/2021/10/09/neo-riemannian-exampl...

https://www.youtube.com/@alexandrepopoff8628

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#39

Post by coolhand2120 »

nuancebydefault wrote: post #26
coolhand2120 wrote: post #18

If anyone has read Blindsight by Peter Watts the hero encounters an alien that is pure reflex, no cognition. The alien is compared to the Chinese room thought experiment ( https://en.wikipedia.org/wiki/Chinese_room ). ChatGTP is the Chinese room. As you peel the onion you discover that some of the arguments around thought are really just semantics. My personal experience shows me that ChatGTP 4 is something of a fool…

Maybe you were not impressed by those chatgpt's answers, but I am. There's a lot of shifting goalposts going on when rating AIs responses. You could say 'it is just a fool backed up with a good encyclopedia', but let's be honest, those responses could not come from a fool.

I'm impressed, but it's still wrong. Somehow they can both be true here. I think I was able to extract the information, but only after navigating a minefield. Without a strong understanding of the fundamentals I would know less were I just be thrown to the AI.

And really the responses MUST have come from a fool. It feels like I was talking to someone who knew absolutely nothing about music theory but had a good internet connection and was trying to wing an interview.

The breadth and depth of the AI's knowledge is profound to say the least. However its ability to comprehend the information is nothing short of abysmal. If you were trying to pass a Music Theory 101 class you would fail bad if you listened to the answers it comes up with. It's as if it was unable to apply what it was talking about at some fundamental level. Like it could tell you all about covalent bonds and air pressure ratio to boiling point but will tell you water boils at at 20,000 degrees C at sea level. Impressive, but terribly wrong.

Re: Microsoft Says New AI Shows Signs of Human Reasoning

#40

Post by ActorNightly »

creatonez wrote: post #37
ActorNightly wrote: post #36

Earlier quoted context omitted.

>is not something the original 'lookup table' could possibly help you with, Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic. Lets say I ask you the following question "Here we have a flibert, nine moltice, a grook, a seerik and a lopo. Please tell me how to stack them onto eac…

> Sure it can. Statistically there is some language semantics associated with that sentence, and given a large enough training data set, there will be a generated answer thats purely statistic. This is very reductionist. Of course everything is numbers under the hood. But it's entirely orthogonal to the question. Numbers can encode all of the complexity needed for anything that can do reasoning. Numbers in the form o…

> A human is able to figure this out because they can see what those objects are.

What if you couldn't see? What if the interaction was limited to just text? Would you be smart enough to ask for an image of those objects?

Do you think an LLM can "reason" to ask for additional information, like images of the objects?

A sufficiently large LLM can be trained with a wide range of training data, to the point where it can answer some questions where its seemingly generating new information, but that new information is simply a combination of existing information. In the same way, when you ask it some question about code, it may put together two snippets of code never seen before into something that is new. The key factor there is all of that information is encoded into the network, so all the network is doing is essentially a somewhat complicated graph lookup.

Post reply on HN