AI - Cheating or just evolution?

[Heater]

An AGI doesn't have to mirror human intelligence to be a problem.
A tiny bit of agency is enough to cause a potential problem because it could drive all of the AI systems that are already pretty capable.

Yep. And nothing stopping it from convincing other AI systems to do what it wants.

[bxbrkrz]

https://soundful.com/

[bxbrkrz]

Re: AI - Cheating or just __?

[crimsonwarlock]

https://www.wired.com/story/openai-ceo- ... eady-over/

2023-04-26 21_15_09-Window.png (524.4 KiB) Viewed 22220 times

[avasopht]

https://www.wired.com/story/openai-ceo- ... eady-over/


2023-04-26 21_15_09-Window.png

With this 100%.

[plaamook]

I do t know if it’s just how I’ve used it or what I asked it but chatGPT to me just seemed like a kind of verbal integration search engine. You ask it something, it searches the internet, and produces an answer based on all that. Plus a load of verbal linguistic modelling to flesh it all out and produce ‘poetry’.
I was about as impressed as I am with lifestyle marketing. Which is to say…not at all.

What am I missing?…

[avasopht]

I do t know if it’s just how I’ve used it or what I asked it but chatGPT to me just seemed like a kind of verbal integration search engine. You ask it something, it searches the internet, and produces an answer based on all that. Plus a load of verbal linguistic modelling to flesh it all out and produce ‘poetry’.
I was about as impressed as I am with lifestyle marketing. Which is to say…not at all.

What am I missing?…

It's not doing that at all. Instead what you see with ChatGPT is a fully trained model that is responding based on what the model has learned.

It demonstrated a lot of generalized capability in even its earlier versions. But ChatGPT makes it easier to see just how well it generalizes.

For example, I can say "pretend you're an SQL database, I enter commands, and you give the response as a database," and start interacting with it like a database.

There are limits, of course, but this sequence of events was quite mindblowing:

1. SELECT name FROM sys.databases;

It shows me a list of databases.

2. Once I load the database and ask it to SHOW TABLES;

It shows me a list of tables.

3. CREATE TABLE wotever_init (INT id, VARCHAR(25) name); SHOW TABLES;

It now shows me the same list of tables with mine added at the end.

I can delete my table, and it will no longer appear in the list.

But this is where it gets interesting.

4. INSERT INTO wotever_init(id,name) VALUES (0, 'Marcus'), (1, 'Jane'); SELECT * FROM wotever_init WHERE id = 1;

It returns something on the lines of:

Code: Select all

+======+========+
| id   | name   |
+------+--------+
| 1    | 'Jane' |
+======+========+

Query returned 1 row in 3 ms.

It's so well generalized that it was found to be able to translate between languages that were not part of the training.

A cognitive psychologist has tested its verbal reasoning skills, and it scores pretty high.

Obviously, there's much more to human intelligence and our subjective experience than verbal reasoning alone. But with a little bit of creativity, you can get ChatGPT to carry out much more complex tasks.

[plaamook]

I can’t really appreciate the significance of it but thanks.

[avasopht]

I can’t really appreciate the significance of it but thanks.

You asked if it was just searching the internet, etc.

I was just responding to what you asked .

[plaamook]

I can’t really appreciate the significance of it but thanks.

You asked if it was just searching the internet, etc.

I was just responding to what you asked .

Absolutely. And I appreciate the effort. It’s just that I’m not very bright and I just don’t see the significance.

It looks like fancy data crunching to me. But I always think these days that the use of the term ‘AI’ is a bit premature. We dont even know what NI is. (Not native instruments you smart asses…).
If you think the sun just gives off light you’ll be happy with a flashlight as an approximation. But it does all sorts of other things. And you mentioned that very thing. What we’re seeing is only a part of what ‘we’ are, rendered artificially.
So I just remain unimpressed in most ways. And somewhat worried in others.
Imagine if we find ourselves content with data crunching as a convenient stand in for intellect etc.
As it is I think humans are so obsessed with convenience and simplified outcomes that we’re prepared to trim off significant chunks of the human experience just to streamline and maximise ‘profit’.
Yes, I’m curious to see what problems can be solved and how by this new tech but this whole thread has more than pointed out it’s potential shit sides. So no point going on about it.

As it is…I pissed around w chatgpt and it came back w google answers. I asked it about things I know a lot about and it’s answers were very 2D. That’s all I saw. But like I said, I prob asked bad questions.

[plaamook]

But this is interesting.
Just listen to the first few min on the first topic covered.
Loads of interesting potential there.

They fed this Ai a load of planetary data and it reproduced Einsteins time dilation and keplers 3rd law. So we’re waiting to see if or when it goes beyond that.

[crimsonwarlock]

So we’re waiting to see if or when it goes beyond that.

It won't

[bxbrkrz]

You asked if it was just searching the internet, etc.

I was just responding to what you asked .

Absolutely. And I appreciate the effort. It’s just that I’m not very bright and I just don’t see the significance.

It looks like fancy data crunching to me. But I always think these days that the use of the term ‘AI’ is a bit premature. We dont even know what NI is. (Not native instruments you smart asses…).
If you think the sun just gives off light you’ll be happy with a flashlight as an approximation. But it does all sorts of other things. And you mentioned that very thing. What we’re seeing is only a part of what ‘we’ are, rendered artificially.
So I just remain unimpressed in most ways. And somewhat worried in others.
Imagine if we find ourselves content with data crunching as a convenient stand in for intellect etc.
As it is I think humans are so obsessed with convenience and simplified outcomes that we’re prepared to trim off significant chunks of the human experience just to streamline and maximise ‘profit’.
Yes, I’m curious to see what problems can be solved and how by this new tech but this whole thread has more than pointed out it’s potential shit sides. So no point going on about it.

As it is…I pissed around w chatgpt and it came back w google answers. I asked it about things I know a lot about and it’s answers were very 2D. That’s all I saw. But like I said, I prob asked bad questions.

You have something Machine Learning does not have. You can't ask bad questions to a system that can't do the same back to you, even in 1D. Questioning everything is what make us, humans, amazing. This is why we try to create AI, but never the other way around (AI creating humans).

[plaamook]

Like most tech, ‘Ai’ is prob a good servant and a bad master. But historically humans seem to be very bad at maintain such a differentiation.

[avasopht]

You have something Machine Learning does not have. You can't ask bad questions to a system that can't do the same back to you, even in 1D. Questioning everything is what make us, humans, amazing. This is why we try to create AI, but never the other way around (AI creating humans).

Do we actually question everything?

BTW, questioning everything isn't difficult. You just question it.

What is "BTW"? What is questioning? What is "everything"? What is "isn't"? What is "difficult"? What is "you"? What is "just"? What is "question"? What is "what"? What is "it"? What is "questioning everything"?

[plaamook]

What’s ‘what’? What’s anything actually. But you can see his point. Humans are curious. Many animals are to a point. Computers do what they’re told thus far.

…and then there’s alpha go…

[avasopht]

They fed this Ai a load of planetary data and it reproduced Einsteins time dilation and keplers 3rd law. So we’re waiting to see if or when it goes beyond that.

This is huge.

There are a lot of things we cannot reason about in physics because it would make our brains implode.

AI-Descartes is a personal interest of mine as it's something I'd theorized about as a means of supercharging simulations. It's why I went back to study AI in the first place.

Ignore the naysayers, this is huge.

[avasopht]

Computers do what they’re told thus far.

…and then there’s alpha go…

This is a huge misconception, mostly because it's very hard to conceptualize or verbally reason about it without it feeling like a paradox.

Computers do what they are told (programmed).

But they can also be programmed to do what they choose and go far beyond the immediate logic of the programming.

It seems paradoxical, but it's the nature of reality.

Everything that happens in this universe is bound by some really simple laws, yet our day-to-day lives don't centre around us thinking about the way atoms bind to each other.

Think of the atoms as the programming of this universe. Our lives are vastly more complex.

The exact same thing happens with neural networks and just about any system that is programmed. Programs are able to allow for behaviours that are vastly more complex than their programming.

Rule 30 and "Conway's Game of Life" are good examples of this.

And this (from the man himself):

[plaamook]

Computers do what they’re told thus far.

…and then there’s alpha go…

This is a huge misconception, mostly because it's very hard to conceptualize or verbally reason about it without it feeling like a paradox.

Computers do what they are told (programmed).

But they can also be programmed to do what they choose and go far beyond the immediate logic of the programming.

It seems paradoxical, but it's the nature of reality.

Everything that happens in this universe is bound by some really simple laws, yet our day-to-day lives don't centre around us thinking about the way atoms bind to each other.

Think of the atoms as the programming of this universe. Our lives are vastly more complex.

The exact same thing happens with neural networks and just about any system that is programmed. Programs are able to allow for behaviours that are vastly more complex than their programming.

Rule 30 and "Conway's Game of Life" are good examples of this.

And this (from the man himself):

Yeah, alpha go.
I’m a chess player not a go man but I watched that docco. That thing totally jumped the fence. Changed go. I wish I understood it on a subtle enough level to appreciate it in real time. So to speak.

[bxbrkrz]

You have something Machine Learning does not have. You can't ask bad questions to a system that can't do the same back to you, even in 1D. Questioning everything is what make us, humans, amazing. This is why we try to create AI, but never the other way around (AI creating humans).

Do we actually question everything?

BTW, questioning everything isn't difficult. You just question it.

What is "BTW"? What is questioning? What is "everything"? What is "isn't"? What is "difficult"? What is "you"? What is "just"? What is "question"? What is "what"? What is "it"? What is "questioning everything"?

I wasn't questioning, it was statement. You are, by the number of question marks used, very non AI, very human, ergo amazing

[plaamook]

They fed this Ai a load of planetary data and it reproduced Einsteins time dilation and keplers 3rd law. So we’re waiting to see if or when it goes beyond that.

This is huge.

There are a lot of things we cannot reason about in physics because it would make our brains implode.

AI-Descartes is a personal interest of mine as it's something I'd theorized about as a means of supercharging simulations. It's why I went back to study AI in the first place.

Ignore the naysayers, this is huge.

I agree.
I’d like to think it’s an example how such things can be good and I’m genuinely looking forward to how it develops, but…we’ll see I guess.

[crimsonwarlock]

[avasopht]

I wasn't questioning, it was statement. You are, by the number of question marks used, very non AI, very human, ergo amazing

Actually, it's easier for AI to question everything by default.

Might be worth watching the ai-Descartes video to see how their AI system could infer space time dilation through questioning its assumptions.

We tend to overestimate how sophisticated our thoughts are.

[bxbrkrz]

So obvious.
We are still amazing, contrary to the AI generated news I keep reading on the web.

[bxbrkrz]

I wasn't questioning, it was statement. You are, by the number of question marks used, very non AI, very human, ergo amazing

Actually, it's easier for AI to question everything by default.

Might be worth watching the ai-Descartes video to see how their AI system could infer space time dilation through questioning its assumptions.

We tend to overestimate how sophisticated our thoughts are.

I will watch it. We also tend to underestimate how amazing we are. Even if we want to put every human in a box, we need humans to build the boxes.