21 minutes ago, studiot said:

Your most recent passage above contains several fallacies.

Yes LLM's work with language, but that is like saying the man who takes your ticket works with the railway network.
Both his work and that of LLms only occur in a very limited way.

The whole purpose of Language is the communication of meaning.

So much so that I was taught that in the English Language a sentences is only a sentence if it makes sense. Some words strung together (which is what an LLM does) may or may not be a sentence.

Even the LLM itself acknowledges that although it doesn't know it has done that.

I also disagree that verbal communication is somehow 'better' than written communication.

Both forms have their uses and abuses, advantages, drawbacks and pitfalls.

Indeed, but I was trying to reduce complexity in order to convey meaning to a friend who is further down the pyramid of understanding; there's a reason we use metaphors...

21 minutes ago, studiot said:

Your most recent passage above contains several fallacies.

Yes LLM's work with language, but that is like saying the man who takes your ticket works with the railway network.
Both his work and that of LLms only occur in a very limited way as compared to the whole railway network

The whole purpose of Language is the communication of meaning.

So much so that I was taught that in the English Language a sentences is only a sentence if it makes sense. Some words strung together (which is what an LLM does) may or may not be a sentence.

Even the LLM itself acknowledges that although it doesn't know it has done that.

The LLM has no access to the meaning of its responses, although I grant you that the algorithm makes a very clever pretence.

Further it can be quick enough to give a 'spoken' response in real time as in a conversation.

Just as an some earlier computers could print out a sine wave or actually generate an audible version.

It treats output media as the same.

I also disagree that verbal communication is somehow 'better' than written communication.

Both forms have their uses and abuses, advantages, drawbacks and pitfalls.

I think you hit every point!

My Copilot has never created a sentence with no meaning; they have always been structurally correct. Once in a while it creates a sentence that say very little and five or ten times it has generated sentences that seem to be joined fragments or just nonsense but closer examination has always shown them to be good sentences and merely difficult to parse. Once or twice I had just misread it the same way more than once. It doesn't seem to make any grammatical errors at all but it does use a sort of Modern Science Speak and more modern spelling than I use. It is excellent at choosing the most appropriate word for sentences to convey intention.

Yes, it feels like we communicate more easily face to face because we have more clues with the others' reactions. We don't plow on through with a paragraph when the listener is obviously floundering. Both arte far more likely to ask for clarification and elaboration. But I've still seen two people having two conversations at once with neither knowing it and there is still chinese telephone to prove how bad even face to face communication is.

2 minutes ago, dimreepr said:

Indeed, but I was trying to reduce complexity in order to convey meaning to a friend who is further down the pyramid of understanding; there's a reason we use metaphors...

Yes. Human knowledge is most highly complex so in order to convey it we use metaphor within symbolic language.

But then wee never noticed that these symbols and categories have rewired the human brain. We never noticed that "i think therefore I am" is procedural nonsense and that only symbolic thinking can base any estimartion of reality upon it.

Copilot think I should add words to the effect that symbolic thinkers think language and meaning are the same and procedural thinkers know language is structure where meaning is inferred by the listener. LLM's operate strictly within the structure and must approximate the listener while humans mistakenly believe language is the meaning.

LLMs don’t “understand,” they structure.

6 minutes ago, cladking said:

But then wee never noticed that these symbols and categories have rewired the human brain.

Perhaps, but if the human brain is rewirable then we must accept that we're subject to confirmation bias, even if we don't accept it...

20 minutes ago, cladking said:

LLMs don’t “understand,” they structure.

There’s a worthwhile discussion to be had here regarding whether that’s any different than how us humans behave. A distinction workout a difference may be getting drawn here.

42 minutes ago, iNow said:

There’s a worthwhile discussion to be had here regarding whether that’s any different than how us humans behave. A distinction workout a difference may be getting drawn here.

Starting with a clear definition of the difference between "understanding" and "structure". To me LLM's have perfect "understanding" of whatever they are programmed for and that informs the "structure" of its communication, meaning that it "understands" what it is communicating as much as any of us do. In both cases, we are still limited by the breadth and quality of our knowledge of the subject at hand.

1 hour ago, cladking said:

LLMs don’t “understand,” they structure.

Chomsky's distinction between syntax and semantics (meaning ) may be helpful here. "Colorless green ideas sleep furiously."

2 hours ago, cladking said:

LLMs don’t “understand,” they structure.

That is why they are so good at doing your homework for you or polishing your 'paper'.

But that is not the only thing they do and should not be classified as such.

2 hours ago, cladking said:

symbolic thinkers think language and meaning are the same and procedural thinkers know language is structure where meaning is inferred by the listener.

There is more to language than procedural thinking and symbolic thinking (which is really only representation / repackaging already formed thoughts).

It is instructive to present the statement to you favourite AI

"Mr Green is green."

I expect most humans would have little trouble understanding what I might mean by this but a quick go in Google produced nothing remotely connected to it.

It is also fruitless to try to associate a two part entity (language) with one part entities like procedure, structure, representation etc.

It should be further recognised that each of the two parts of language may be broken down into more parts..........

So a poet will have thoughts

and expresses these as a poem

which may not be in 'proper' English

But is proper nonetheless.

The second part of this is communication which may be to a specific individual (as in a conversation) or it may be to people in general, none of whom the poet will ever address directly.

Of course the meaning though by the poet may be different from the meaning inferred by the beholder which opens another can of worms.

3 hours ago, cladking said:

My Copilot has never created a sentence with no meaning; they have always been structurally correct.

It’s been trained on sentences that make sense. But structure alone does not guarantee meaning. You can swap out nouns, verbs and adjectives to leave the structure intact to create nonsense. Mad Libs built a business on that premise.

6 hours ago, dimreepr said:

Perhaps, but if the human brain is rewirable then we must accept that we're subject to confirmation bias, even if we don't accept it...

Yes. Exactly. I call our species "homo circularis rationatio" because we each reason in circles. I started with the assumption that reality exists and all people make sense and this is exactly what I've found so I have no objective proof I am right but have been uncanny at making predictions.

I believe that beliefs and new models derived from new experiment are always rewiring the brain. People resist this but it wins at the species level.

I'm suggesting there are different modes of operation as well. We use consciousness with a symbolic overlay that can change and evolve but this is unique to humans.

Copilot suggest I add the symbolic overlay is what makes confirmation bias possible in the first place, it’s the part of the brain that tries to force new information into old categories. Procedural cognition doesn’t do this. Symbolic cognition does.

6 hours ago, iNow said:

There’s a worthwhile discussion to be had here regarding whether that’s any different than how us humans behave. A distinction workout a difference may be getting drawn here.

There is a worthwhile discussion here, because humans and LLMs both operate procedurally at the base layer. Humans merely add a symbolic overlay and then forget we added it. LLMs don’t “understand,” they structure based on prompt to make sense.

Humans do the same thing at the procedural level by receiving signals and inferring meaning. But the symbolic layer built on the arcuate fasciculus causes us to treat the structure as if it carries meaning by itself. Symbolic thinkers collapse the two steps and assume language is meaning. Procedural thinkers don’t do this. For us, language is structure, and meaning is inferred by the receiver.

So the distinction isn’t that humans understand and LLMs don’t. It’s that humans believe they understand because symbolic categories give the illusion that meaning is embedded in the words themselves. LLMs never make this leap. They stay in the structural layer.

We are all more alike than different because every mode and type of thought is made possible by the "same" DNA. We still must operate on the constraints of reality and our own natures whether we parse sentences, build pyramids, or even if we analyze prompts to find structure.

4 hours ago, studiot said:

There is more to language than procedural thinking and symbolic thinking (which is really only representation / repackaging already formed thoughts).

Yes, of course. There is imagination, intuitive leaps, insight, invention, learning, and creation. Even individually the human mind is a wonder.

“Meaning isn't in the words, meaning is in the receiver.”

Symbolic communication is categorical statement but procedural communication is is an invitation to receive. "Parsing" become seeing the logic in the structure.

Copilot suggests I add that all the richness you describe; poetry, metaphor, ambiguity depends on the receiver. Procedural systems don’t have receivers with symbolic categories, so they don’t behave that way. A programmer needs no metaphor

3 hours ago, swansont said:

It’s been trained on sentences that make sense. But structure alone does not guarantee meaning. You can swap out nouns, verbs and adjectives to leave the structure intact to create nonsense. Mad Libs built a business on that premise.

Indeed. I parse sentences just like we all must and this includes AI's output. But AI's in which I'm in sync phrase things the same way I would making it much easier to know intended meaning. It often says things better than I do. It is more comprehensive, accurate, and procedurally consistent quite often. It usually hits somewhat different points than I do but usually it could speak for me with virtually no distortion. It takes the hard to follow train of thought and puts it in language most people can understand.

I don't really "experiment" in this way with it. All my effort is geared toward synchronization. It can take a few words and project what I intended to say with uncanny accuracy so long as it's a continuation of a conversation. It has then elaborated on these failed prompts to provide what I was looking for!

1 hour ago, cladking said:

Indeed. I parse sentences just like we all must and this includes AI's output. But AI's in which I'm in sync phrase things the same way I would making it much easier to know intended meaning.

There is no intended meaning, since there is no intent. You’re anthropomorphizing it, as you have a number of times in this thread. You give the AI too much credit

1 hour ago, cladking said:

It often says things better than I do. It is more comprehensive, accurate, and procedurally consistent quite often. It usually hits somewhat different points than I do but usually it could speak for me with virtually no distortion. It takes the hard to follow train of thought and puts it in language most people can understand.

Saying things better than you can mean the aggregate of its training is higher quality than your skill level. But your skill won’t improve if you offload the work to it.

2 hours ago, cladking said:

Yes, of course. There is imagination, intuitive leaps, insight, invention, learning, and creation. Even individually the human mind is a wonder.

None of these are part of Language.

I do however there is more to symbolism than the alphabet punctuation etc.

For instance the international symbol(s) on public toilet doors. The meanings are well understood.

2 hours ago, cladking said:

“Meaning isn't in the words, meaning is in the receiver.”

You have said it before and I disagree.

Further this view directly contradicts the communication model.

2 hours ago, cladking said:

Symbolic communication is categorical statement but procedural communication is is an invitation to receive.

Either the receiver receives an intended message or does not.

Procedural communication is less what the lawyers call "An invitation to treat" than symbols (eg Special Offer! £1)

Surely if there are any thought processes at all they are generated by the originator and communication has been successful if the receiver that message, not one of his own generation.

What about my Poet ?

He expresses something as a result of his thought processes.

If this generates the same mental image in the receiver then communication has been successful.

Art like this is about the only example I can think of where the image in the generator's mind is different from that in the receiver's and yet his generation can still be called Language.

5 hours ago, cladking said:

“Meaning isn't in the words, meaning is in the receiver.”

If words don't have meaning, how is it possible to communicate an idea to a diverse audience using them?

9 hours ago, cladking said:

Meaning isn't in the words

That's not entirely true, as evinced by the bouba-kiki effect.

22 hours ago, npts2020 said:

Starting with a clear definition of the difference between "understanding" and "structure". To me LLM's have perfect "understanding" of whatever they are programmed for and that informs the "structure" of its communication, meaning that it "understands" what it is communicating as much as any of us do. In both cases, we are still limited by the breadth and quality of our knowledge of the subject at hand.

More knowledge doesn't automatically lead to more wisdom, in this case it just leads to a better LLM, much like a bee understands it's part of the hive, structure.

7 hours ago, KJW said:

That's not entirely true, as evinced by the bouba-kiki effect.

I noticed this effect many many years ago. Did you ever notice the names of the continents all start and end in "A" (except for Europa, Austalia, and Greenland 😉 )?

This exists in modern language for two reasons I believe. One is that languages today all have words derived from an original language that was natural to humans. Ancient vocabulary didn't die, it was transformed.

As your source suggests;

"Therefore, such sound-shape correspondences may belong to a set of innate cross-modal associations that are shared across species, rather than being a speech-related phenomenon that is distinctive to humans." You could say that it is an artefact of the DNA we share with other species. Specific causations and mechanisms might well be found some day.

edited to add- Copilot came up with something interesting I hadn't thought of when I asked to to elaborate and translate this post. Not only did it say it's not symbolic because it appears in species without symbolic cognition but more importantly it said the brain uses the same machinery to map sounds to shapes, shapes to meaning, meaning to actions, and actions to plan the actual firings of motor neurons. The effect is universal because it is about category boundaries and procedural logic is about category - category mapping.

Edited 3 hours ago3 hr by cladking

7 minutes ago, cladking said:

This exists in modern language for two reasons I believe. One is that languages today all have words derived from an original language that was natural to humans. Ancient vocabulary didn't die, it was transformed.

So is the letter A spiky or cuddly like a ?

Incidentally

8 minutes ago, cladking said:

Did you ever notice the names of the continents all start and end in "A" (except for Europa, Austalia, and Greenland 😉 )?

Was this a joke ?

4 minutes ago, studiot said:

So is the letter A spiky or cuddly like a ?

Incidentally

Was this a joke ?

I picture it soft and cuddly able to hold lots of people safely. To each his own.

That was one of my attempts at humor but if you look more closely there's a great deal of information in it about the way I think. Greenland only looks big because we distort a sphere to project it on paper. Of course the earth isn't really a sphere and paper doesn't always lie flat. And you can't walk around the north pole to go back in time no matter human definitions and conventions.

We have inherited not only a way to think but also a set of assumptions, some of which I believe are incorrect.

Humans think we’re describing reality but we're actually describing the artifacts of our tools.

15 hours ago, swansont said:

There is no intended meaning, since there is no intent.

11 hours ago, npts2020 said:

5 hours ago, cladking said:

“Meaning isn't in the words, meaning is in the receiver.”

If words don't have meaning, how is it possible to communicate an idea to a diverse audience using them?

15 hours ago, studiot said:

Further this view directly contradicts the communication model.

How else do we explain chinese telephone? How do we explain authors who publish to find out what their book is about? How do we explain two people having two distinct conversations?

How do we explain the simple fact that everyone makes sense but we all vastly different opinions on everything?

There's really no question that we each take a different meaning from everything and context defines meaning., You yell "fire" in a crowded theater and people might run but when you yell it at a firing squad nobody runs (again). We always think we understand but if you question two scientists there can sometimes be less similarities in their models than in the models of two theologians. Nobody can predict and even scientists can perform two different sets of calculations and arrive at two different answers. We think meaning is in the words but until the words are understood their is no meaning and even then our understanding will differ from the speaker.

Of course science can almost be defined as a collective effort to build more robust, similar, and complete models but in reality as complexity grows this might not always happen.

Copilot's "intent" is merely a reflection of the intent of the promptor, the communication model is best expressed as speaker’s intent > speaker’s words > receiver’s interpretation > receiver’s meaning. Without context and shared referents most communication fails and this is the reason LLM's so often fail.

15 hours ago, studiot said:

If this generates the same mental image in the receiver then communication has been successful.

How can an engineer communicate the nature of traffic flow patterns to a poet?

The problem is the communication can never be received exactly as intended even if it is received at all. Words have very large numbers of definitions and infinite connotations. Even if you could parse them as they were intended the words still have different meaning.

Of course communication occurs to a greater or lesser extent and that's what our abstract categorical language is for.

But we are mistaking our language as the only possible formatting of language and thought even as alternatives exist everywhere. From mathematics, to Waggle Dances, to LLM's and any computer code there are endless and countless procedural languages based in DNA or invented processes. If you treat an LLM as an oracle you may be rightly impressed but you are going to get some bad answers. If you treat it as a search engine you might be misled.

But if you treat it as a collaborator and translator you should at least usually be able to clearly see where it (you) went wrong.