Those claiming AI training on copyrighted works is “theft” misunderstand key aspects of copyright law and AI technology. Copyright protects specific expressions of ideas, not the ideas themselves. When AI systems ingest copyrighted works, they’re extracting general patterns and concepts - the “Bob Dylan-ness” or “Hemingway-ness” - not copying specific text or images.
This process is akin to how humans learn by reading widely and absorbing styles and techniques, rather than memorizing and reproducing exact passages. The AI discards the original text, keeping only abstract representations in “vector space”. When generating new content, the AI isn’t recreating copyrighted works, but producing new expressions inspired by the concepts it’s learned.
This is fundamentally different from copying a book or song. It’s more like the long-standing artistic tradition of being influenced by others’ work. The law has always recognized that ideas themselves can’t be owned - only particular expressions of them.
Moreover, there’s precedent for this kind of use being considered “transformative” and thus fair use. The Google Books project, which scanned millions of books to create a searchable index, was ruled legal despite protests from authors and publishers. AI training is arguably even more transformative.
While it’s understandable that creators feel uneasy about this new technology, labeling it “theft” is both legally and technically inaccurate. We may need new ways to support and compensate creators in the AI age, but that doesn’t make the current use of copyrighted works for AI training illegal or unethical.
For those interested, this argument is nicely laid out by Damien Riehl in FLOSS Weekly episode 744. https://twit.tv/shows/floss-weekly/episodes/744
Citation needed. I’m pretty sure LLMs have exactly reproduced copyrighted passages. And considering it can created detailed summaries of copyrighted texts, it obviously has to save more than “abstract representations.”
If I asked you to recite a popular poem, nursery rhyme, a song, or book passage there’s a good chance you could. Everyone can recite things word for word.
It’s the same with LLM’s, if they’re asked to generate, for example, an article written by the New York Post about a specific topic they really did write about, then it’s similar to asking someone to recite a poem or song.
Not many. And generally not book passages or whole NY Post articles. That’s the point. OP claims it tosses the original, but it doesn’t.
Yes, literally every single person on this planet can recite a song or poem.
But there are naturally massive differences between a human brain and an LLM. The point I was making is that an LLM doesn’t copy and store books and articles wholesale. The ability to reproduce samples from the dataset is more of a quirk than a feature, in the same way that a person can memorize things.
But that is just it. When a commercial enterprise is literally saving copyrighted content and car reproduce it on demand, copyright holders have every right to object. Either use public domain materials and/or license copyrighted materials, or don’t try to make money off AI.
Where is the LLM that can reproduce specific whole copyrighted works on demand? All ive seen is reproductions of quotes of a few sentences (fair use) and hacks that can make it ocasionally vomit up random larger fragments of its training data, maybe up to a few paragraphs.
LLMs do it in a commercial setting.
And? No one said otherwise. The comment I was responding to made the argument that LLMs merely memorize content, which isn’t true.
I didn’t say that at all. I was responding to OP claiming they don’t memorize content at all.
You’re right, I’ve responded to a few comments here and I thought it was another comment thread I was replying to.