Artificial Intelligence and Music

[jeremyb]

Thought I'd share some of the things that excite me in AI but are related to music rather than the more general things I'm playing with.

This is a really interesting talk using generative AI as a creative tool:

[hamo]

A friend of mine is a DJ and he reckons there's already some amazing AI tools integrated into the software he uses. Exciting times.

[werdna]

Software to write ACDC songs.

https://www.youtube.com/watch?v=vpEVsDN84Hc

The same every time. Perfect.

[Mr Bungle]

AI created cover of Paul McCartney "performing' Aha's Take On Me. I think the backing track is from an Aha Unplugged performance.

[bender]

I use a handful of noise reduction tools that are built around neural networks. Pretty amazing stuff.

[calling card]

Is AI smart enough to know how to switch off the power and reboot? My HP Design Jet printer can't.

[jeremyb]

Uh oh...

[jeremyb]

Interestingly a year on I now think AI is the new blockchain, moronic senior managers everywhere are wanting AI added to everything to fix problems that don't exist, but thats ok the massive power and computing costs will be covered by making more developers redundant...

[hamo]

I think it's more that ChatGPT etc has skewed the story about what it is, what it could be useful for, and how it needs to be implemented. But you have the dual issue of it being sold as a magic box that does everything, so that's what uneducated business people want it to be, so we'll see lots of terrible implementations; but also it's the latest fairy dust that shareholders demand you shoehorn into your product, so there's even more impetus for those bad implementations. But because it's easier for certain people to wrap their heads around "magic box go brrrr" than blockchain was, you'll find it more pervasive, at least for a while, as it uses the same amount of power as small central American countries annually...

[jeremyb]

One amusing thing is most of the large language models are close to running out of source material, all that human generated content that they feed upon, things are going to hit the wall very soon…

[hamo]

The more useful applications will be the ones that are trained on more constrained, better curated sets of source material, then pointed at specific use cases. The more general AI just has too many opportunities to go rogue.

[Cdog]

IMG-20240610-WA0000.jpg

[jeremyb]

The more useful applications will be the ones that are trained on more constrained, better curated sets of source material, then pointed at specific use cases. The more general AI just has too many opportunities to go rogue.

Imagine how confused it would be if all you gave it was ACDC albums

[hamo]

The more useful applications will be the ones that are trained on more constrained, better curated sets of source material, then pointed at specific use cases. The more general AI just has too many opportunities to go rogue.

Imagine how confused it would be if all you gave it was ACDC albums

[sty]

Interestingly a year on I now think AI is the new blockchain, moronic senior managers everywhere are wanting AI added to everything to fix problems that don't exist, but thats ok the massive power and computing costs will be covered by making more developers redundant...

The underlying tech of the current AI wave is really interesting, but like you say it's being massively mis-interpreted (as usual).

Back in the mid 90s I worked with some amazing Neural Net technology from a Cambridge University spin-off company (the one that was bought by HP and was stuck in legal battles with HP recently). It was truly amazing, especially for the time.

I was using it in some of the very very early E-Commerce sites in Europe, Great Universal Stores Catalogues, I trained it on their entire catalog (the old phone directory sized catalog things you're all familiar from growing up) just the short and long written descriptions of all their products. It was amazing at working out what content was important and made things unique and could suggest alternatives with incredible accuracy, this was incredibly helpful to a catlog company that ran out of stock and needed to suggest substitutions.

The tech today is built on the kind of same technology but way bigger, it can't genuinely do original thought. but man can it find the nuanced stuff you want it to.

proper Machine Learning is amazing, see above, but also it was being used in the late 90s to detect credit card and financial fraud because it could identify patterns and more importantly wrong/bad patterns.

Today it's uses are amazing, it can do face recognition incredibly well, I've seen examples in NZ of it being used to distinguish between pests/predators and endagered species and being used in traps to poison pests and alow valuable species to pass unharmed. In the world of photo editing it can make incredibly powerful tools to crop out background, to fill in voids, all very useful stuff, to unblur photos, even to colourise very old photos and sun damanged photos. It can do the same for video and in the audio world it will be able to perform miracles, imagine being able to remove tape hiss, rescue terrible mixes, improve the quality of live sound etc. These are just tools, not creative and will make a huge difference.

Likewise in medicine, the ability to process X-Rays, Mamograms, MRI scans etc. will save many lives.

What todays AI really isn't is intelligent, and it certainly isn't really capable of creative thought, yep it can probably create the illusion of creativity usually through the tiny amounts of chaos present in systems, but it's a trick, it's not human etc. By the same token a lot of humans aren't capable of creative thought/generation either, so it can likely fake it for a while.

A lot of companies and niaive managers are going to lose a lot of money and make bad choices based on the current and next gen of AI if they think they can use it as a replacement for their creative humans. Yep a few are going to get lucky and score big, but a hell of a lot more are going to quietly run up huge bills, make huge cock ups, and quietly move on to the next big thing.