Software vs Hardware

[BONES]

And yet they do exactly that every time you use them. I think you are confusing "knowledge" and "understanding". They sort of know everything but understand nothing, just like people. That said, they really do understand a lot, otherwise they'd be useless. If I ask for something more aggressive, it understands what I want and is able to produce it. If I ask for lyrics using iambic pentameter, it understands what I am asking and tries to comply. That it can't always manage it makes it all the more interesting, in that it is able to balance different understandings - the words need to make sense, they need to fit the brief but they also need to fit the meter - which shows a kind of deeper understanding, that of competing priorities. Sometimes it decides that making sense and/or fitting the brief is more important than sticking rigidly to the meter. It's understanding may not come about in the same way as yours or mine but it is a form of understanding nonetheless.

[pdxindy]

And yet they do exactly that every time you use them. I think you are confusing "knowledge" and "understanding". They sort of know everything but understand nothing, just like people. That said, they really do understand a lot, otherwise they'd be useless. If I ask for something more aggressive, it understands what I want and is able to produce it. If I ask for lyrics using iambic pentameter, it understands what I am asking and tries to comply. That it can't always manage it makes it all the more interesting, in that it is able to balance different understandings - the words need to make sense, they need to fit the brief but they also need to fit the meter - which shows a kind of deeper understanding, that of competing priorities. Sometimes it decides that making sense and/or fitting the brief is more important than sticking rigidly to the meter. It's understanding may not come about in the same way as yours or mine but it is a form of understanding nonetheless.

That's just you anthropomorphizing it.

You start off saying it understands nothing, and by the end of your paragraph, you have convinced yourself it has some form of understanding.

Using human terms like "understanding" for AI, is about the same as ascribing human characteristics to your mirror because your reflection mimics your movements so well.

The mirror and AI have an equal capacity for knowledge and understanding.

[Tj Shredder]

AI is huge hardware combined with huge software, both not residing in my studio. Is that the context which belongs to that topic?
I‘d rather discuss that in Off Topic. That ping pong of exactly the same non-arguments is a bit boring…

[BONES]

That's just you anthropomorphizing it.

That's just you thinking you're special. You're not. Animals, even stupid ones, demonstrate both knowledge and understanding, they are not uniquely human traits. Therefore it makes no sense to suggest any level of anthropomorphising.

You start off saying it understands nothing, and by the end of your paragraph, you have convinced yourself it has some form of understanding.

You need to learn to read better. I said "sort of", which leaves plenty of wiggle room.

Using human terms like "understanding" for AI, is about the same as ascribing human characteristics to your mirror because your reflection mimics your movements so well.

So when you tell your dog to sit and he obeys, it has no understanding? When a cat learns how to operate a door handle to get into a room with a closed door and is able to repeat that behaviour over and over, there's no knowledge or understanding there, either? Don't be ridiculous.

[crimsonwarlock]

I think you are confusing "knowledge" and "understanding".

You think wrong, I'm not confusing anything here.

They sort of know everything but understand nothing, just like people. That said, they really do understand a lot, otherwise they'd be useless.

Nope. They don't know anything and they don't understand anything. LLMs are built on top of Transformers (Google tech), which in turn are built on top of Deep Learning systems which are, in fact, Artificial Neural Networks. ANNs store pattern statistics, that's it. The only information that is trained into a LLM is the statistical occurrence of a word together with a cluster of other words. Your prompt is translated to the cluster values of the words in your prompt, and the LLM matches this statistically to the best cluster of words for an answer. Knowledge, understanding, or anything even close to that, has absolutely nothing to do with how LLMs work.

Simply put, LLMs do not store any semantic information. The tech doesn't work that way, no matter how you perceive the results.

[El°HYM]

[muzicxs]

Nope. They don't know anything and they don't understand anything.

You are absolutely correct. LLMs are all about output token prediction and have nothing do to with intelligence.

[markello]

Comparing LLMs to dogs is rough.

[IvyBirds]

Nice example of what software can do with a nice programmed MIDI controller

[BONES]

They don't know anything and they don't understand anything.

Don't be ridiculous, it knows and understands way more and way better than you or I. Ask it a question and it will give you the answer. It won't always be correct but it will be right a helluva lot more often than any human.

LLMs are built on top of Transformers (Google tech), which in turn are built on top of Deep Learning systems which are, in fact, Artificial Neural Networks. ANNs store pattern statistics, that's it.

As opposed to what? All you're telling me is how they work, it proves (or disproves) nothing.

The only information that is trained into a LLM is the statistical occurrence of a word together with a cluster of other words.

And how do you think your brain works? It's called a "neural" network for a reason. It works in the same way our brains do. It's just a lot more thorough than we are.

Simply put, LLMs do not store any semantic information. The tech doesn't work that way, no matter how you perceive the results.

So what? Again, you're not proving or disproving anything. You can have an intelligent conversation with it. It can't make the same kinds of intuitive leaps that we can but it can certainly hold up its end of any conversation you care to have with it. That requires both knowledge and understanding. How it arrives at that understanding, be it sophisticated algorithms or neural networks or whatever, is irrelevant.

It sure as hell understands more than most feeble, little minds. You just have to look at how good they have gotten at producing release-quality music in the last year or so to see that. It's pretty much indistinguishable from human-created stuff these days. Does that disturb you? Why? It's just another tool for us to leverage. It can be a million times smarter than me and I wouldn't be worried or intimidated by it, there is no reason to be.

You are absolutely correct. LLMs are all about output token prediction and have nothing do to with intelligence.

Well, the Oxford Dictionary defines intelligence as "the ability to acquire and apply knowledge and skills". How does AI not fit that definition? We all know that it acquires data centres full of knowledge, it's "training", and if you interact with it, it is obvious that it is able to apply that knowledge in useful, meaningful ways. So how is that not intelligence?

Comparing LLMs to dogs is rough.

I think you mean "ruff!".

[cryophonik]

Comparing LLMs to dogs is rough.

Because its Bard is worse than its byte?

[pdxindy]

Nice example of what software can do with a nice programmed MIDI controller

It has digital oscillators, and everything past that is analog (filter, wavefolder, distortion, etc.)

[IvyBirds]

Nice example of what software can do with a nice programmed MIDI controller

It has digital oscillators, and everything past that is analog (filter, wavefolder, distortion, etc.)

No it's not, that Demo is all digital as is indicated in the description of the video itself. The entire sound engine in the demo is all digital


Maybe you missed that part? Here is the screen shot

Screenshot_20250513-215425.jpg

Every sound you hear in that video is software, and the physical controls are just controllers for that software.

[vurt]

Comparing LLMs to dogs is rough.

ruff...

[markello]

Comparing LLMs to dogs is rough.

I think you mean "ruff!".

Comparing LLMs to dogs is rough.

ruff...

Good to see my carefully crafted post was appreciated:)