ilovefuzz.com

I haven't been on the up and up but I heard Soundcloud started doing this last year, and that's the one I was using most recently. Is Bandcamp doing this too?

As in the company themselves using user content to train AI models, or it being possible for others to scrape data?
Bandcamp AFAIK hasn't done anything like this (yet).

It is possible to scrape any track with a preview stream from Bandcamp without paying, so theoretically anything you upload could end up as training data, but because preview streams are all 128kbps mp3, they would not be very useful.
Companies will do any number of nefarious things to circumvent actually paying for training data, so it's a real concern, but it's high quality (in this case high-bitrate, extensively/accurately tagged) data that they're after.

backwardsvoyager wrote: ↑Sat May 31, 2025 2:44 am As in the company themselves using user content to train AI models, or it being possible for others to scrape data?
Bandcamp AFAIK hasn't done anything like this (yet).

It is possible to scrape any track with a preview stream from Bandcamp without paying, so theoretically anything you upload could end up as training data, but because preview streams are all 128kbps mp3, they would not be very useful.
Companies will do any number of nefarious things to circumvent actually paying for training data, so it's a real concern, but it's high quality (in this case high-bitrate, extensively/accurately tagged) data that they're after.

The former, companies training their AI on artist uploads. I don't make music for the money, but I also don't exactly want to voluntarily give companies free reign to make money off of my work while I don't, either.

Bandcamp it is, I guess?

Right, yeah I would give BC the benefit of the doubt. They've done alright by artists even since the buyout.

Not sure about the profit incentive for sites that sell digital DL's, but streaming services are inherently liable to start using AI in efforts to skimp on artist royalities by generating similar content and leading users there via playlist/recommendation algos, etc. (if they haven't already)
I've been following stuff like HarmonyCloak (https://mosis.eecs.utk.edu/harmonycloak.html) as it could well get to the point where we can't upload anything anywhere without it becoming training data, but even then it's hard to say whether poisoning filters, etc. will be a solution.