How popular can AI music become?
I wrote a song using a new prompting method that uses massive context windows in LLMs to understand the Suno and Udio models, and it seems that the music models are much more capable than most people believe - because simply typing in obvious prompts like "pop" yields repetitive, flat audio.
"Rythmos Bay" is 100% generated by AI models (but stitched together from multiple generations.) Initial reviews by people on another site suggested that people believed it was the best AI song they had heard, a sentiment that was shared by in-person listens. On Manifold, however, reviews have been more polarized.
Every other day until the end of November, I will post the current number of listens the song has received since its release on November 14, a few hours before the market was opened. The song was initially simply posted in comments as a .flac file, and has now been published to YouTube. I will continue to publicize this song over the next few days as I have time to post it to other sites.
The resolution of the market will be the number of listeners, summed across all sources. I'll post the sources where the song is currently hosted and update the market as those new sites are added. I'll also update the number of listens from each source. If the song gets a large number of views, then the "Other" category will be split to add new categories as the existing limits approach.
This market can also be seen as an observation on whether AI outputs can be effectively "marketed," regardless of the quality of the output. In some cases, I may omit mentioning that the song has no human performances in it.
CURRENT SOURCES:
YouTube: https://www.youtube.com/watch?v=mMV3LALOJmY (67)
Original direct download: https://shoemakervillage.org/steve_sokolowski-rythmos_bay.flac (77)
SoundCloud: https://soundcloud.com/steve-sokolowski-797437843/12-rythmos-bay (18)
X: https://x.com/SteveSokolowsk2/status/1857073106474385889 (does not count in total because it is a link to other sites)
Bandcamp: https://stevesokolowski.bandcamp.com/track/rythmos-bay (0)
TOTALS BY DATE:
November 14, noon EST: 38
November 15, 8am EST: 74
November 30, 11:59:59pm EST: 162
1,000
3.00The answer to the market is 162. The breakdown is in the comments.
If the song gets a large number of views
Are you going off of listens or views? Can you distinguish between the two?
@ian I don't think it would be fair to distinguish between the two, because each platform counts different things.
YouTube, for example, provides deep statistics and I can see that some people didn't listen to the end. The webserver logs for the master files just show whether a person downloaded the song. On other sites, you can only see whether people started listening.
The definition of a listen will be "having at least one byte of the song downloaded to the user's computer."
@GG I was thinking of creating a "radio edit" version if this version was listened to, so that's a good idea.
But 2:40 wouldn't even be long enough for a bridge. I do see how I could create a version that's about 4:10 and still be complete.
Are you suggesting there is some place that it would gain more exposure if these cuts were made?