Please watch this video of Udio being used to generate output based on metadata describing Bladee's music.

Then listen to Bladee's music: open.spotify.com/album/0cT1SQDE…

What's the training data?

If a new gallery opened that you had reason to believe stole the artworks it displayed, you wouldn’t write about it without mentioning this.

Journalists, if you’re writing about a generative AI product and the signs point to it being trained on copyrighted work without…

People are uploading music to the DSPs that is generated using AI products that don’t reveal their training data, and that therefore may be trained on copyrighted work without permission.

This means it will compete with music it may be trained on.

How To Make GREAT Content!

Copyright laundering in action.

If company A trains a model on creators' work without permission, and company B generates synthetic data for their own model using company A's model, company B is exploiting creators' work without permission, too.

As Stephanie Palazzolo says, this…

A reminder that the general public don’t buy the argument that, just because humans can learn from publicly available work, AI models should be able to.

In the largest public poll on this question I’ve seen (2,052 people in the US), when people were asked how AI companies should…

Generative AI products usually collect ratings and feedback on output from users (👍/👎 etc.). This is generally called preference data. They use it to improve their models, and it makes a big difference.

If their models are trained on copyrighted work without permission, this…

Today’s Daily Cartoon, by Zoe Si. #NewYorkerCartoons

Janne is another AI music researcher who *actually* respects musicians.

Remember that the opinion that all creative output is there for the taking by gen AI models is not universally held among AI researchers. There are many, like Janne, who understand the impact of ignoring…

neil turkewitz Stuart Dredge Absolutely these should precede just about anything. If it’s not there within the core value system of the company, I have a very hard time believing Udio will end up being a “net good for the industry”.

PSA: if you get output from a gen AI model that suggests it was trained on copyrighted work without permission, always download it + record the prompt you used.

If you don’t, the service you generated it on can delete the evidence.

Ed Newton-Rex Worrying that prompt suggestions are straight-up actual album names:

Udio has launched, and musicians will have serious questions regarding what it’s trained on.

Answering a question on this, Udio say they need lots of inputs, they deflect to the question of exact regurgitation, and they use the terms ‘transformative’ and ‘publicly available’.…

The Generative AI Copyright Disclosure Act introduced by Adam Schiff is great, but there's one change it would be great to see: require gen AI companies to disclose *all* their training data, not just copyrighted works.

The main reason would be to ensure synthetic data is…

Here's the full text of Adam Schiff's proposed Generative AI Copyright Disclosure Act. A couple more interesting points:

- What's required to be disclosed is 'a sufficiently detailed summary of any copyrighted works used'
- There will be a public database of these disclosures…

MONEY

“The best musicians do not feel threatened by [AI music]” - this is demonstrably false.

Just last week over 200 of the best artists and songwriters in the world wrote a letter that directly refutes this.

We need to put an end to the lie that it is only ‘mediocre’ artists who…