The Music Thread

beefchopper said:
I find this pretty scary. At least it should scare the hell out of musical artists. This song was composed and entirely produced and "recorded" from a single request to AI.


Click to expand...
See, I’m neither dismissive or apocalyptic when I hear these types of AI created songs.

I’m not dismissive because the technology is clearly there to displace many of what I would call the “lowest common denominator” types of music, or music that is very pattern based, or has predictable structures, where an LLM generally excels. So, think of things like stock music, jingles, advertisements, EDM, or any background music requiring a general mood but not nuance. AI will destroy the economics of these businesses for songwriters, who would otherwise be hired for these projects. So musicians should be very concerned there. It's not going away.

However, I’m not apocalyptic either for two reasons. One, many people tend to anthropomorphize AI these days, but LLMs are essentially fancy, sophisticated statistical calculators rather than sentient creators. They can excel with the aforementioned music, but an LLM is likely not giving us the next Sinatra any time soon. A lot of the nuance, phrasing, and depth is not something that I think an LLM will ever really convincingly generate and it’s one reason why I always find these examples technically impressive but creatively bland, flat, and lifeless. I’m not sure that can be fixed.

Two, and maybe most importantly, many of these companies are likely training these models on stolen IP, and the lawsuits are already coming in from the RIAA and the major labels. We’ll see how “fair use” is defined, but I suspect there will be enough to slow these companies down or force them to have to exclude copyrighted material from their training data, which itself could impact the quality of their models.

Honestly, if I were an aspiring musician, I’d be more concerned about labels continuing to use AI to increasingly automate key roles like A&R to a machine. I think that’s a bigger threat right now, but that’s a separate issue. Either way, we’re in the Wild West with all this right now. It’s not quite the same, but these music generating services remind me quite a lot of the early Napster days. It’ll be interesting to see how this plays out.
 
Matthew said:
See, I’m neither dismissive or apocalyptic when I hear these types of AI created songs.

I’m not dismissive because the technology is clearly there to displace many of what I would call the “lowest common denominator” types of music, or music that is very pattern based, or has predictable structures, where an LLM generally excels. So, think of things like stock music, jingles, advertisements, EDM, or any background music requiring a general mood but not nuance. AI will destroy the economics of these businesses for songwriters, who would otherwise be hired for these projects. So musicians should be very concerned there. It's not going away.

However, I’m not apocalyptic either for two reasons. One, many people tend to anthropomorphize AI these days, but LLMs are essentially fancy, sophisticated statistical calculators rather than sentient creators. They can excel with the aforementioned music, but an LLM is likely not giving us the next Sinatra any time soon. A lot of the nuance, phrasing, and depth is not something that I think an LLM will ever really convincingly generate and it’s one reason why I always find these examples technically impressive but creatively bland, flat, and lifeless. I’m not sure that can be fixed.

Two, and maybe most importantly, many of these companies are likely training these models on stolen IP, and the lawsuits are already coming in from the RIAA and the major labels. We’ll see how “fair use” is defined, but I suspect there will be enough to slow these companies down or force them to have to exclude copyrighted material from their training data, which itself could impact the quality of their models.

Honestly, if I were an aspiring musician, I’d be more concerned about labels continuing to use AI to increasingly automate key roles like A&R to a machine. I think that’s a bigger threat right now, but that’s a separate issue. Either way, we’re in the Wild West with all this right now. It’s not quite the same, but these music generating services remind me quite a lot of the early Napster days. It’ll be interesting to see how this plays out.
Click to expand...
The scary part is the rate of improvement. Given that these AI models are now writing their own new code to improve, I wouldn't discount the threat they make to jobs of all sorts including in the arts.
 
beefchopper said:
The scary part is the rate of improvement. Given that these AI models are now writing their own new code to improve, I wouldn't discount the threat they make to jobs of all sorts including in the arts.
Click to expand...
I find it fascinating that science fiction movies you've seen decades ago come to real life.
 
Matthew said:
See, I’m neither dismissive or apocalyptic when I hear these types of AI created songs.

I’m not dismissive because the technology is clearly there to displace many of what I would call the “lowest common denominator” types of music, or music that is very pattern based, or has predictable structures, where an LLM generally excels. So, think of things like stock music, jingles, advertisements, EDM, or any background music requiring a general mood but not nuance. AI will destroy the economics of these businesses for songwriters, who would otherwise be hired for these projects. So musicians should be very concerned there. It's not going away.

However, I’m not apocalyptic either for two reasons. One, many people tend to anthropomorphize AI these days, but LLMs are essentially fancy, sophisticated statistical calculators rather than sentient creators. They can excel with the aforementioned music, but an LLM is likely not giving us the next Sinatra any time soon. A lot of the nuance, phrasing, and depth is not something that I think an LLM will ever really convincingly generate and it’s one reason why I always find these examples technically impressive but creatively bland, flat, and lifeless. I’m not sure that can be fixed.

Two, and maybe most importantly, many of these companies are likely training these models on stolen IP, and the lawsuits are already coming in from the RIAA and the major labels. We’ll see how “fair use” is defined, but I suspect there will be enough to slow these companies down or force them to have to exclude copyrighted material from their training data, which itself could impact the quality of their models.

Honestly, if I were an aspiring musician, I’d be more concerned about labels continuing to use AI to increasingly automate key roles like A&R to a machine. I think that’s a bigger threat right now, but that’s a separate issue. Either way, we’re in the Wild West with all this right now. It’s not quite the same, but these music generating services remind me quite a lot of the early Napster days. It’ll be interesting to see how this plays out.
Click to expand...
The Japanese have been doing something like this for close to twenty years with their Vocaloid star, Haksune Miku. They’ve even given her a background and a physical presence on TV shows. .”She” does live concerts. I saw one where she was a CG person on stage with a real vocalist. It was easy to tell she was a CG presence but it was impressive for 2010.
 
DodgerSSR said:
We saw Gordon Lightfoot perform live in Houston back in the '80's. I always liked his style, although the better half wasn't a big fan of his.
Click to expand...
Long before I married Herself, I took a certain Chinese paramour to see Gordon Lightfoot perform live. She was bored by his music. That should’ve set off an alarm. Thankfully, her daddy didn’t care for us round eyes. Probably saved me.
 
You must log in or register to reply here.
Back
Top