When I was at school in the 1960s we were told that robots would do that, giving more leisure time. What wasn’t said was that the time freed up would be unpaid (reduced employment) so no money to enjoy the leisure time (or to have a decent living) the benefit going to the owners of the robots.
3 Likes
Isn’t that what Kenny G is already?
2 Likes
What I am hoping for is AI mastering to improve SQ or even replicate some legendary recordings but in the highest sound quality possible.
1 Like
That requires AI to judge what is best SQ better than humans.
I suspect a significant limitation to SQ is not the capability of the mastering engineers ears, but what audience it is being mastered for - some, maybe much at least non-classical, is mastered to sound good to the average person using an average system - which today might be earbuds on a smartphone.
I’ve always been frustrated by this. Some albums are clearly mixed for radio/headphones in noisy traffic.
They should mix them for pristine playback and let modern devices apply a sort of “make it sound like crap EQ”.
AI could be used there. It could determine how much dynamic range compression to apply and recessed midrange and bass boost depending on factors like how many Instagram posts of food have been created, how much Diamond White has been purchased, how often it overhears the user say things like “Bangin’!”, or “Well 'ard” or “Awwright girlfriend!” 
3 Likes
Now dynamically controlled DSP is something where AI might be of value - amongst other things it could make for a truly effective “loudness” control, revolutionising lis level listening
Over the years I’ve bought albums where I’ve found that spark infuriating, annoying, ‘not for me’ etc only to find that with time I tuned in and eventually saw the spark and once hooked I stayed hooked.
Having laid out for an album I felt an incentive to keep trying that I don’t have with streaming and I suppose if people don’t listen to certain AI generated music maybe the algorithm converges everything towards the middle of the road.
1 Like
Are there Beatles fans here?
Who has listened to the last song with AI Lennon voice ? Impressions ?
Now and Then, the long-awaited “final” Beatles song featuring all four members, is to be released next week thanks to the same AI technology that was used to enhance the audio on Peter Jackson’s documentary Get Back.
“There it was, John’s voice, crystal clear,” [Paul McCartney]said in a statement. “It’s quite emotional. And we all play on it, it’s a genuine Beatles recording. In 2023, to still be working on Beatles music, and about to release a new song the public haven’t heard, I think it’s an exciting thing.”
Now and Then was written and sung by [John Lennon]in the late 1970s at his home in the Dakota building in New York City. In 1994, Lennon’s widow Yoko Ono gave the demo to Paul McCartney on a cassette labelled “For Paul” that also included Lennon’s demos for Free As a Bird and Real Love.
While the latter songs were completed by McCartney, Ringo Starr and [George Harrison]) and released as singles as part of the Beatles Anthology project, technological limitations meant that Lennon’s vocals and piano on Now and Then couldn’t be separated to work alongside new parts recorded by the other three Beatles, and it was shelved.
The idea is to have the recording musician(s) or any client get the end product’s SQ right in a relatively short time and expense. Therefore AI facilitates the client input/feedback and offers near instant iterations. There are current plugins for workstations but I expect more intelligent interfaces and algorithms will appear.
The current wave of AI based on LLM:s is not intelligent in the same way as a human (although many of the LLM proponents claim humans are just as dumb as the LLM:s).
The most advanced LLM-project around today is is probably Teslas FSD (self-driving cars) and they are now very close to achieving this (Uber will be replaced by cheap self-driving Tesla RoboTaxi:s). They have been collecting an extremely large amount of training data from all teslas for the last 5-6 years (the cars have 8 cameras plus sensors). So the cars behaviour in traffic comes from all training data completed with knowledge on how fast you can drive before passengers feel scared or physically uncomfortable.
But still they cannot be creative. The solution to all situations must be in the training data which makes it tedious the closer you get to achieving full days of driving without human intervention. And the danger is when the LLM output is being fed in as new training data - the research clearly shows this will make the LLM:s more and more simple-minded repeating the same solution. So one could think the LLM:s have their best time around now and will only become more stupid as they eat more and more of their own output.
Another interesting development is there seem to be a growing number of people who can immediately identify LLM-generated music. I have not seen an explanation.
The alternative in AI is AGI (Artificial General Intelligence). And here the jazz comes in. Or will come in once there is some progress in this area, currently nowhere in sight. Here a self driving car could be given goals and deduce how to behave in traffic without all that training data. And by itself create new rules and goals.
I couldn’t agree less 
Tesla have been a couple of years away from self-driving cars for the best part of a decade now and they’re no closer now than they were then. It’s a running joke on the internet. They have been litigated against repeatedly and were eventually forced to cease describing their vehicles as self-driving. They don’t meet the definition and likely never will. Uber may be replaced at some point by semi-automated taxis but every time such a service has been trumpeted and rolled out so far a series of accidents have resulted in the suspension or withdrawal of said service.
Elements of self-driving will be absorbed into cars as happens now but self-driving itself is as fine example as you could want of something trumpeted that will never happen.