AI Week June 29: Copyright and Bood Food
This hasn't been a happy week for authors who didn't want generative AI trained on their work.
Hi! Welcome to this week's AI week.
This hasn't been a happy week for authors who didn't want generative AI trained on their work. If you don't have time for anything else, just read about the copyright ruling below. But I've included a couple of other things that may amuse you. (Do you like Bood Food? Enjohy!)
In this week's AI Week:
- Copyright
- Something fun
- All bow down to our new AI Lord
- Musk's plan to rewrite the Internet
- Ctrl-Z for art restoration
- Bonus: ELI5 glossary of AI terms
Copyright
Big copyright ruling this week.
Key fair use ruling clarifies when books can be used for AI training - Ars Technica
In landmark ruling, judge likens AI training to schoolchildren learning to write.
The Anthropic copyright ruling reasons that the copies used to train LLMs won't displace demand for copies of Authors' works.
I wish I knew how the judge got there, because that doesn't match my perception and experience.
For example, here's a BookTok influencer advertising a Character AI service to fans to create their own romantasy, using their favourite characters and books, instead of paying for the author's sequels. "Try for free to bring your favorite books and characters to life with immersive storytelling!" That sure sounds like "displacing demand" to me.
However, this ruling says that using copyrighted books to train LLMs is "fair use" as long as the LLM training companies buy the books first, which mostly hasn't happened. Anthropic ripped off a lot of the books it used to train its LLMs, but had a change of heart halfway through and started destructively scanning truckloads of books instead.
Something fun: Bood Food
Is AI responsible for this tray liner?
AI in the wild? This tray tissue I was given at a restaurant is full of errors and hallucinations…!
— Chaz Firestone (@chazfirestone.bsky.social) 2025-06-25T17:46:46.339Z
Worst thing about this tray liner: Some human, somewhere, looked at this and said "Looks good, let's print it." And then other people ordered it for their restaurants. WHY.
All bow down to our new AI Lord (literally)
In the "Humans are weird" department, this week I learned that there are lots of people who believe ChatGPT is divine.
Full disclosure, I didn't make it through all 45 minutes of the video. I cheated and read this article about it instead.
Just something to keep in in mind as you read the next article:
Musk's plan to rewrite the internet
Grok was supposed to be "based", but a large part of its training data was the internet, and the Internet tends to be left-libertarian by US standards. (This may be because the developed world is left-leaning by current US political standards.) Like all other major LLMs trained on the entire Internet, Grok's political leanings reflect that. Musk's plan? Use LLMs to rewrite the entire training corpus to better reflect Musk's biases -- excuse me, I mean, "with better accuracy".
1984, but with LLM’s - by Gary Marcus - Marcus on AI
“Who controls the past,’ ran the Party slogan, ‘controls the future: who controls the present controls the past.’” - 1984, George Orwell
Restoring art reversibly with the help of machine learning
This is cool. Rather than touching up a painting with paint, it's touched up with a printed film that can be removed if the conservator wants to Ctrl-Z the changes and rewind to the base painting.
MIT student prints AI polymer masks to restore paintings in hours - Ars Technica
Removable transparent films apply digital restorations directly to damaged artwork.
The technique doesn't use generative AI, which would distort the image, but does leverage machine learning techniques from computer vision and conservation.
The rewindability is really neat, but I also like learning how fields outside programming are exploring what it could mean to work with machine learning as a "force multiplier," as coders sometimes put it. As Ars says:
In a world where AI models increasingly seem to blur the line between human- and machine-created media, it's refreshing to see a clear application of computer vision tools used as an augmentation of human skill and not as a wholesale replacement for the judgment of skilled conservators.
Bonus: an ELI5 glossary of AI terminology
Do you know your RAG from your RLHF? If your AI vocab could use some fine-tuning, give this post a try.
An AI glossary - by Lenny Rachitsky - Lenny's Newsletter
The most common AI terms explained, simply
That's it for this week's AI week! Thanks for reading.