seeds wrote: ↑Tue Feb 17, 2026 7:39 pm
RickLewis wrote: ↑Mon Feb 16, 2026 1:20 am
I'm stunned. I don't know whether to congratulate you or weep for the future of live music, but in any case that is an outstanding result.
Welcome back, also.
Thank you, Rick, for the welcome back, and especially for your kind and encouraging words regarding my little AI assisted peace anthem:
"We Want Love to Rule the World".
Here's the link to it again for anyone who is interested:
https://suno.com/s/ZZA1rggLzZFoDNir
And again, all clicks on it (even those out of sheer pity) are appreciated.
And yes, by all means, be worried. Besides taking all of our jobs away, AI will soon be wiping our butts, blowing our noses, and making love to our spouses far better than we slow and smelly meat sacks ("non-synthetics") can do any of those things.
_______
This is a useful guide to the difference between what a human assumes is happening when "AI" does its thing and what is actually going on.
When people use AI to generate music, they just kind of assume it does stuff in the same way we do. Just as they assume the AI that writes their homework is composing text in the same basic way we do. But in both cases they are mistaken and that is why the current type of AI model has stopped improving.
When we train AI to write text, we pour millions of books into them in simple text files that are easy to sort and categorise. When promoted to compose a text, the AI recreates some plausible seeming facsimile of those training materials. It doesn't sit and think through an argument if you use it to write a philosophy paper, it assembles at most one sentence at a time as it constructs a stream of text until the engine decides it has reached a stopping point.
Things are, if anything, slightly worse when using AI tools to compose music. They are trained in much the same way, millions of mp3 files are poured into a bucket, accompanied by text describing what humans think the content sounds like, and then when you give it text input describing the music you want, it reaches into that big bag of sounds and assembles something plausible sounding one musical phrase at a time, sort of.
But the mp3 files for rock and roll music are combined. You don't get the Rolling Stones on 4 CDs with one CD for the drummer, and one for the bass. The mp3 is a jumble of many noises all at once and that jumble is what the AI is working with. This explains why, about 3 quarters of the way through your song, there is a sort of almost guitar solo that just fizzles out, but it doesn't really sound much like a guitar, it's a lot like Eddie Van Halen farting through an electric bagpipe. That is the result of training the AI on a pile of compressed noises made out of many sounds all at once.
School children forming a band in their mum's garage are often quite talentless. But no matter how bad they are with those instruments, they understand how to put rock and roll together in simple steps. The drummer lays down a groove, the bass lays something over the top and then some lanky kid with greasy hair complains about his girlfriend getting fingered by a competing gent for up to three minutes, and then they stop. Current AI models couldn't attempt that, and the cost of attempting to feed all the exponentially enlarged data to train it in actual multi-instrumentation would likely be outrageous.
The above also helps explain, I hope, the basics of how the AI composer of your track is throwing the kitchen sink into the mix in a different way each time the song cycles. Not something that any human composer who can conceive of an entire song would do. That aspect is made somewhat more obvious by the song going round far too many times and being much too long for the limited content. But I'm not sure the AI is responsible for that last part. I suspect you encouraged that particular slice of rock and roll excess.
If I were making a Youtube video and I wanted incidental background music, I might well use an AI to compose something bespoke rather than digging through the libraries of royalty-free elevator music that creators on that platform traditionally use. That's what it can make.
When fishpie tried to make AI music a few months ago, he tried to get it to spit out some fiery Nine Inch Nails, but it can't do that, so it gave him some formulaic darkwave technogoth and because he was involved in its creation, fishpie liked it, but I doubt many others did. I don't even object to industrial gothic future pop and I thought it was derivative crap.
I don't know what you prompted the AI to offer you with that song. If you asked for something reminiscent of Swedish Spandex metal-balladeers Europe, then the result was highly plausible. If you asked for Van Halen, then I would say perhaps less so. But I am not that expert at stadium cheese-rock, I see the plastic costumes they wear and instantly suffer a sort of sympathetic jock-itch.