As widely rumoured, the UK government is considering legalising gen AI training on copyrighted work without a licence. This would be a huge betrayal of the UK’s creators. It would mean letting AI companies use creators’ work to build highly scalable AI models that will directly compete with them. This cannot be allowed. All UK creators, and the millions of people who benefit from the brilliant work that comes from the UK’s creative industries, should oppose this. ft.com/content/26bc3de1-af90…

Once you’re familiar with the style of AI-generated text, you see it everywhere. The amount of LLM-generated slop in replies on social media is getting unbearable. LinkedIn is particularly bad. I would much rather read a short, real reply full of grammatical errors than an essay that’s ‘well-structured’ and perfectly spelled but that represents the meaningless average of all possible thought.

This piece by Ted Chiang is hands down one of the best pieces of writing on generative AI I’ve ever read. He skims over some arguments on both sides of the AI art debate. It will annoy people who see gen AI as one creative tool among many, just as it will annoy people who think the question of whether AI can make art is a distraction from the real-world exploitation -> job displacement cycle inherent to much gen AI. But for its take on push-button gen AI, and for the sheer quality of the writing, for me it’s unparalleled. Cannot recommend reading it highly enough. newyorker.com/culture/the-we…

It's time for streaming services to act on AI music. - AI companies are training on musicians' work without a license - People are making music with these models and putting it on streaming services - This music is being recommended to users This means human musicians' royalties are being impacted by AI models trained on their music against their wishes. Streaming services should label AI music, and either ban or downweight music made using AI trained on unlicensed data in recommendations & royalty calculations. If they continue to sit on the fence, I suspect we'll start seeing musicians pulling their music from these platforms in protest. Read my piece in @musicbizworld today: musicbusinessworldwide.com/i…

Two Midjourney engineers once told me art was “ossified”, and used this to justify scraping people’s work for training data. They said art had stopped progressing, and these models were what would take it forward. This betrays, I think, not only a lack of respect for artists, but also a lack of understanding of art. Pleased to see this lawsuit move forward.

Some crazy stuff in this ProofNews update on Nvidia’s mass scraping of YouTube etc.: - They say in internal messages that there’ll be no negative sentiment because they won’t reveal their training data - Employees raised concerns, those concerns were dismissed - They internally justify scraping by saying it’s for research purposes, despite these being commercial models - They scraped copyrighted videos in part because they think OpenAI are doing the same (proof that the arms race is driving these actions) - They cite fair use in their statement to ProofNews, essentially admitting the reports are accurate AI company employees who raise concerns about training on copyrighted work, only to be ignored, should quit. proofnews.org/nvidia-scrapes…

The 3 major record labels are suing AI music companies Suno and Udio. Here are the two lawsuits in full. - They accuse Suno & Udio of “willful copyright infringement on an almost unimaginable scale” - They provide evidence that both companies trained on their music, including outputs that closely resemble their recordings (ABBA, Michael Jackson, Green Day, James Brown, & many more) - They outline why this is not fair use - They say this “wholesale theft of… copyrighted recordings threatens the entire music ecosystem and the numerous people it employs” - They include unknown co-defendants who assisted in copying/scraping - They demand a jury trial If you do one thing today, read the full complaints. 🧵 of some of the outputs 👇 Suno: s3.documentcloud.org/documen… Udio: s3.documentcloud.org/documen…