Master AI (Artificial Intelligence) Discussion/News Thread

[pablolie]

I get what you are saying, but personally I’ve also really appreciated/enjoyed generative music, going back to analog and computer experiments by Eno (or earlier avant-gardists) or heavily auto-tuned dance/pop from Charli xcx.

Also - how will you establish when/if your fav artist needs a hit and leverages AI to generate tunes? Is it just wrong when an utterly unknown scores a hit? The established name has the "credibility" to get away with it?

 

[Rick63]

I get what you are saying, but personally I’ve also really appreciated/enjoyed generative music, going back to analog and computer experiments by Eno (or earlier avant-gardists) or heavily auto-tuned dance/pop from Charli xcx.

Heck, I am one of those that hated when synthesizers came into prominent use in the 70's, and really hate electronic drums. I love Rush and they're one of my favorite bands, and they lost me completely after Permanent Waves.

 

[Rick63]

Also - how will you establish when/if your fav artist needs a hit and leverages AI to generate tunes? Is it just wrong when an utterly unknown scores a hit? The established name has the "credibility" to get away with it?

Even worse is that many of those "artists" are winning Grammy awards. But then, the Grammy Awards have had about as much credibility as the Rock and Roll Hall of Fame has had for many years, like sometime late in the 20th Century...

 

[Axo1989]

You are of course free to argue. Personally I just see a trend of bullsh*t, not a new culprit. It's still humans using a tool to fake skills they don't really possess. Nothing new. Like a driver with a car that makes over 350hp that is just driveable by electronic controls. Pure posing. Nothing new. …

That’s an impressive zero-to-car-analogy time you managed there.

… Come up with that line when "AI provenly kills people". It's just lazy sensationalism otherwise. …

Hence ‘currently litigated’.

… Not AI's fault when people use the wrong tool for the wrong purpose. Read the NOT fine print undr every AI prompt. You can blame the company. You can blame humans. AI algorithms themselves are quite stupid and don't create any of these issues unless prompted to do so by unqualified people …

Well AI is a science fiction concept describing something not invented yet. And a misconstrued marketing term for a grab-bag of generative tools. I’m not blaming either for outcomes, but you are still arguing as if outputs are deterministic.

 

[pablolie]

...
Well AI is a science fiction concept describing something not invented yet. And a misconstrued marketing term for a grab-bag of generative tools. I’m not blaming either for outcomes, but you are still arguing as if outputs are deterministic.

AI has been around for a LONG time in computing discussions. I first encountered it in University in the 80s when discussing "Expert Systems", which led with algorithmic intelligence rather than a huge data lake that could be abstracted into helpful rules.

You could say that real intelligence is based on algorithmic complexity. What we call "AI" in this discussion is repetitive (and actually pretty dumb) parallel processing of a huge data lake via a small number of relatively simple neural algorithms that are optimized for specialized functions... but -again- leveraging a HUGE data lake. No more R programming crap.

Newton supposedly observed *one* falling apple falling on his head and derived a formula out of it through a lot of math diligence to model the experience. AI would collect 10 billion events of apples falling on people's heads and -if you used it properly- help you correlate those events. It is an immense difference.

 

[Axo1989]

Also - how will you establish when/if your fav artist needs a hit and leverages AI to generate tunes? Is it just wrong when an utterly unknown scores a hit? The established name has the "credibility" to get away with it?

Not sure that ‘hits’ are the right paradigm for me, even in the Charli xcx case I enjoyed the initial singles that later became the album Brat some time before that latter was released/hyped. But my discovery behaviours are probably idiosyncratic.

And I’ve previously posted Arca’s deliberate and overt use of the generative tool Bronze to ‘generate tunes’.

Much deployment of entirely generated music so far has been covert but sufficiently clumsy to easily recognize as sloppy Spotify playlist-stuffing and algorithm-gaming. But yours is a good question. I guess personally I just check out new artists in some depth before forming judgement, assuming I like what I hear in the first place.

 

[sam_adams]

That's an example for an IT department where no one should have any job anywhere near IT.

They were lucky that the company that does their backups is as good at restoring data as they are at backing it up.

 

[pablolie]

They were lucky that the company that does their backups is as good at restoring data as they are at backing it up.

Luck shouldn't have anything to do with it. Even minimally applying ITIL recommended basic practices would have completely avoided that utterly idiotic outcome.

 

[MattHooper]

If I was king of the world I would require the AI monopolies to use IA to cure cancer before being allowed to do anything else with it.

I hereby vote for you as king!

(“you don’t vote for kings!…)

 

[pablolie]

Why would anyone think AI as is can solve issues human experts can't? AI knows nothing and can't learn nothing beyond what human experts know. And with current algorithms never will.

 

[MattHooper]

Why would anyone think AI as is can solve issues human experts can't? AI knows nothing and can't learn nothing beyond what human experts know. And with current algorithms never will.

Even experts are surprised by AI’s latest ‘vibe-mathing’ advance

A ChatGPT AI has proved a conjecture with a method no human had thought of. Experts believe it may have further uses

www.scientificamerican.com

I would say one of the most perilous predictions is what AI can or can’t do, or will or will not be able to do anytime soon…
(even AI experts have been caught off guard by what AI has managed to do and how fast it has advanced).

 

[MattHooper]

i have a 1 year (free) subscription to Perplexity. every now and then, I use it to *try* to troubleshoot issues with my Linux based puters.
every now and then, I ask it "WHY ARE YOU TRYING TO BREAK MY SYSTEM?"
which gives me the answer, "You're right, I'm sorry. I should have explained better / warned you / take precautions (etc) before giving you that instructions".
if "pleasing" the end user means giving bad or dangerous instructions to do something at this point of time we are ,I honestly believe the AI bubble will burst.
investors are looking for profit and money retursn as we speak, AI companies are looking how / where to squeeze its "addicted" users for subscription money. it's just a matter of time.
sure, for creating very specific "things" - it is awesome and beyond. common sense and true logic? I don't see it happen

I seem to we’ve been rather lucky than since AI has been very helpful, solving all sorts of issues.

Here’s one from among many:

My father-in-law who lives in a retirement home, started having a problem where he wasn’t getting sound anymore from his computer, including playing the Internet.

He had a bit of a tricky set up (set up by his son) with outboard, amplification outboard speakers. But he is pretty much legally blind, and was helpless to fix it.

Normally, I would’ve gone over to troubleshoot, but I wasn’t in the physical condition to do so.

So my poor wife had to go over and try and help him. You could not point to anybody less technically inclined in the world than my wife.

When she phoned and described the complicated set up, I just said it could be any number of different things so just turn on ChatGPT use the voice mode and have it guide you through some troubleshooting.

She did that and it helped her solve the problem within 5 to 10 minutes.

Recently ChatGPT has helped me troubleshoot and solve some network issues, some browser issues etc.

I admit, I scratch my head when people say they can’t find any use for it.

 

[MattHooper]

I get what you are saying, but personally I’ve also really appreciated/enjoyed generative music, going back to analog and computer experiments by Eno (or earlier avant-gardists) or heavily auto-tuned dance/pop from Charli xcx.

Am I reading you correctly and that you are open to the possibility of enjoying fully AI generated music?

That’s something I’m certainly struggling with.

At the moment, it gets a huge no from me.
I certainly wouldn’t want to deprive other people of what they enjoy listening to.
But for me, I’ve just always maintained a view and a feeling that when I’m experiencing art of any kind I’m connecting to an artist, and so my appreciation is directed to an artist.

When AI creates songs, just like when it creates artistic photos or paintings or movie clips, I feel the ground beneath my feet shift. Now there’s art but no artist. I don’t know how to aim my appreciation.

And for me the role of somebody doing prompts does not inspire that appreciation.

I don’t know maybe it’s like learning to understand and appreciate rap music.
To a lot of people outside rap it just looked like rap were stealing bits and pieces of other musicians work - those musicians did the hard work of coming up with those songs and hooks, and the rap musician is just piggybacking off of that. Again, at least for quite a long time that was a perception many people had of rap (I admit I still feel tinges of this, even though I have appreciated rap). But then many people were won over once they became more familiar with rap.

So maybe that will happen with AI music. I don’t know. But for now I find the prospect deeply unnerving.

 

[AA63]

Why would anyone think AI as is can solve issues human experts can't? AI knows nothing and can't learn nothing beyond what human experts know. And with current algorithms never will.

Odd take. The greatest benefit of (super) computation or programming that continuous learns from itself and outside information - is being able to run problems/simulations rapidly, so human "experts" or anyone else for that matter, can ask "what if we try" or "what if we look at it this way" - the same as scientists do, just faster, and/or continuously. Having the ability to associate (programmatically via instruction or freely) is a strength of AI and that alone can, has, and will solve existing problems. Humans will have to conduct experiments to verify/validate etc., but I don't think (by that comment) that you fully understand what these tools can do.

 

[pablolie]

Even experts are surprised by AI’s latest ‘vibe-mathing’ advance

A ChatGPT AI has proved a conjecture with a method no human had thought of. Experts believe it may have further uses

www.scientificamerican.com

I would say one of the most perilous predictions is what AI can or can’t do, or will or will not be able to do anytime soon…
(even AI experts have been caught off guard by what AI has managed to do and how fast it has advanced).

flat article. doesn't quote "experts" just kinda. makes it up selectively.

 

[pablolie]

Odd take. The greatest benefit of (super) computation or programming that continuous learns from itself and outside information - is being able to run problems/simulations rapidly, so human "experts" or anyone else for that matter, can ask "what if we try" or "what if we look at it this way" - the same as scientists do, just faster, and/or continuously. Having the ability to associate (programmatically via instruction or freely) is a strength of AI and that alone can, has, and will solve existing problems. Humans will have to conduct experiments to verify/validate etc., but I don't think (by that comment) that you fully understand what these tools can do.

wrong. it relies on constant training. when used according to user guides training users who are supposed to be power users. outside of that it's just idiots juggling chain saws.
every prompt has a warning to not blindly trust results... Come on.

 

[AA63]

wrong. it relies on constant training. when used according to user guides training users who are supposed to be power users. outside of that it's just idiots juggling chain saws.
every prompt has a warning to not blindly trust results... Come on.

Like I said, I don't think you understand/know enough. You might be thinking of ChatGPT or one of the commercial front-ends for LLMs.

It does not "rely" on constant training, it can function with any body of knowledge/memory, which it already has a vast amount of. The constant updating is of raw data/information, and the tuning/programming of how it deals with it, is of course the owners/company that is running its choice.

So again, these are odd takes. And none said to blindly trust anything (in fact I specifically addressed) so quoting the boilerplate disclaimer doesn't mean much.

 

[pablolie]

Like I said, I don't think you understand/know enough. You might be thinking of ChatGPT or one of the commercial front-ends for LLMs.

It does not "rely" on constant training, it can function with any body of knowledge/memory, which it already has a vast amount of. The constant updating is of raw data/information, and the tuning/programming of how it deals with it, is of course the owners/company that is running its choice.

So again, these are odd takes. And none said to blindly trust anything (in fact I specifically addressed) so quoting the boilerplate disclaimer doesn't mean much.

The body of memory you renamed is called a KV Cache in inference (AI in actual action), the steps forward in a transaction with an LLM are called tokens. Be careful cailing yourself an authority when it is clear you know little. AI completely and totally relies on constant training and supervision, that is 101 AI basics. Take a basic Nvidia course, they are free.

 

[AA63]

The body of memory you renamed is called a KV Cache, the step,forwards in a transaction with an LVM are called tokens. Be careful cailing yourself an authority when it is clear you know little. AI completely and totally relies on constant training and supervision, that is 101 AI basics.

LOL and now you're being presumptuous. I'll leave this back and forth here, it is benefiting no one as we are clearly talking in different directions.

 

[pablolie]

LOL and now you're being presumptuous.

No I am stating the very basics any AI practitioner calling themselves halfway competent should know.

AI is useless without a LOT of initial training by humans, and continued training and supervision afterwards while in operation.

Claiming anything else is complete ignorance of AI basics.