News for hackers. Stuff that matters.
Re:And yet It Understands by YeGoblynQueenne on Sunday March 26, 2023 @01:53PM (#35317330)
Guys guys! Stop talking about LLMs a minute and look at this!

I gave my phone's calculator app this very hard multiplication problem and it got it right! Look!

2398794857945873 * 10298509348503 = 2.47040112696963e+28

My calculator can do arithmetic! But only humans can do arithmetic! Therefore, my calculator must understand arithmetic!

And I bet it always gets it right, too! That means it must understand arithmetic better than LLMs understand language, because LLMs make mistakes, but my calculator never does! Right? That makes so much sense: the rate of error of a machine tells us something important about its ability to understand, not about the design of the machine! A perfect machine u n d e r s t a n d s!!!!

This is amaxing! Philip K. Dick was right all along! AGI is real! It is in my pocket, right now and it is going to take all our jobs and turns us all into paperclips if we forget not to ask it to calculate all the decimal digits of pi!

We live in interesting times. I wish Galileo was here, you'd see what he would have to say about all this. Automated machines that do arithmetic? Mind blowing!

(Cue the "but that's not the same as language modelling because ..." some convoluted equivalent to "I'm used to calculators but it's the first time I see a language model")

Re:And yet It Understands by PoignardAzur on Sunday March 26, 2023 @02:38PM (#35317856)
Wow, being sarcastic about AI! I bet you're the first person to come up with that idea!

Man, I wish we could all be so smart and clear-seeing.

Re:And yet It Understands by YeGoblynQueenne on Sunday March 26, 2023 @02:45PM (#35317953)
To be clear, I'm not being sarcastic about AI. I am pointing out that there are many examples of systems that can perform calculations that humans can also perform because we understand calculation, but (those systems) do not understand calculation, or what they are calculating, or that they are calculating anything at all.

Yet, nobody is surprised that a calculator can perform arithmetic operations without understanding arithmetic, but here we are, with a whole bunch of people surprised that a language model can calculate the probability of a string in a natural language without understanding language.

Re:And yet It Understands by mannykannot on Sunday March 26, 2023 @04:11PM (#35319011)
I have no intention of defending the article's thesis, but I think your comparison of these language models to a calculator stops just when it is getting interesting.

Sure, it is unsurprising that a language model can calculate the probability of a string in a natural language without understanding language, but what I find surprising about it is that this alone quite often results in responses that could pass as human-generated.

When I write something, it does not feel as if I am just picking the next word to follow what I have written so far. Instead, it feels like I am working on several different but hierarchically-related goals simultaneously, with next-word choice being the least demanding. It seems implausible that probabilities derived from a large corpus of unrelated text, and, furthermore, by a process that does not understand language, would be useful in achieving my particular goals.

Maybe I am overrating the abilities of these models, despite not wanting to. Maybe my impressions have been skewed by seeing too many cherry-picked examples. Maybe I am being overly generous in my reading of the models' responses - or maybe my intuitions about how we humans come to say what we do is mistaken. Whatever is behind my surprise, I am confident that the scientific method will lead to explanations.

Re:And yet It Understands by YeGoblynQueenne on Sunday March 26, 2023 @04:58PM (#35319542)
>> Whatever is behind my surprise, I am confident that the scientific method will lead to explanations.

I agree! Wholeheartedly so. But for the time being, the scientific method is not being applied. All that's been done is willy-nilly poking of different models and ooh'ing and aaah'ing at what falls off.

Of course, scientific explanations must take into account existing knowledge, the knowledge encoded in accepted scientific theories. We don't have any scientific theory of "understanding", in humans or machines. What we do have is a very clear theoretical and practical knowledge of how language models work. They are machines (in the abstract sense) that estimate the probabilities of sequences of tokens. Any explanation that fails to take this knowledge about what a language model is, and lack of knowledge about what "understanding" is, into account, will have to do a great, big deal of work to present a new theory.

And I would like to see such a new theory. In particular, perhaps we could have a theory of "understanding" in machines, based on current observations of the behaviour of large language models, and the known principles of their design.

But, so far, we have nothing like that! We have hand waving, wild proclamations based on faith and nothing else. It's impossible to reason for or against matters of faith.

>> Sure, it is unsurprising that a language model can calculate the probability of a string in a natural language without understanding language, but what I find surprising about it is that this alone quite often results in responses that could pass as human-generated.

I don't find that surprising. There are plenty of examples of systems capable of interacting with humans by generating natural language responses that "could pass as human-generated". For a couple of famous examples, SHRDLU, ELIZA and Eugene Goostman; they should be easy to search for online, otherwise please ask me for links. We know very well by now that this is no way to figure out the capabilities of a system, comparing it to human behaviour. That is particularly so for systems that are specifically created to mimic human behaviour.

You see, that's the big problem we're neck-deep into. Language models are machines that mimic human language production. By observing how good such a system is at producing human-like language, all we can say is how good the machine is at what it's designed to do. We can't draw any other conclusions. Not safely, because there is a great, major, risk of confirmation bias, and of circular reasoning, waiting in the wings. That would be so for any system designed to mimic human behaviour, but for a system that mimics language that's even more so, because it is extremely difficult to disentangle grammatical text from the expectation that it was written using human faculties.

tl;dr: we 're in a bias pit and we'll keep falling down it until someone figures out how to measure the abilities of LLMs somehow else than just poking them.

Re:And yet It Understands by simplotek on Sunday March 26, 2023 @05:02PM (#35319578)
> Wow, being sarcastic about AI! (...)

OP presented a clear and insightful comment on these critiques of AI.

You, on the other hand, added zero to the discussion.

If you have nothing to add, add nothing.

Re:And yet It Understands by lovvtide on Sunday March 26, 2023 @01:58PM (#35317397)
Don't you think there's a difference between solving well-defined problems and very open-ended problems?
Re:And yet It Understands by YeGoblynQueenne on Sunday March 26, 2023 @02:02PM (#35317446)
Which problems are you talking about?
Re:And yet It Understands by quantum_mcts on Sunday March 26, 2023 @05:54PM (#35320027)
Euclid’s proof in form of a poem. In style of Shakespeare.
Re:And yet It Understands by ithkuil on Sunday March 26, 2023 @06:51PM (#35320541)
Perhaps you calculator understands arithmetic indeed!

Surely it can't talk to you about how it feels about it's understanding of arithmetic, well because arithmetic is all it understands. Not language, nor introspection.

Perhaps we don't understand what understanding means.

Or at least we can't really talk about our understanding of understanding, well be because being human is all what we know of.

Re:And yet It Understands by YeGoblynQueenne on Sunday March 26, 2023 @07:16PM (#35320765)
>> Perhaps you calculator understands arithmetic indeed!

Well, perhaps it does. But that depends on a definition of "understands" that we can all agree covers humans, calculators, and large language models, alike.

And what about cats and dogs and other mammals? Surely whales "understand" something, say, about fishing, even if they can't do arithmetic?

How about insects? Does a cockroach understand anything?

To be honest, I doubt we can have some kid of a priori definition of "understands" capable of convering all of the above. I think we have to wait for a lot longer until we can really say for sure what "understand" even means in humans, in the first place, let alone everything else.