Comment by ReflectedImage
2 months ago
All creative types train on other creative's work. People don't create award winning novels or art pieces from scratch. They steal ideas and concepts from other people's work.
The idea that they are coming up with all this stuff from scratch is Public Relations bs. Like Arnold Schwarzenegger never taking steroids, only believable if you know nothing about body building.
The central difference is scale.
If a person "trains" on other creatives' works, they can produce output at the rate of one person. This presents a natural ceiling for the potential impact on those creatives' works, both regarding the amount of competing works, and the number of creatives whose works are impacted (since one person can't "train" on the output of all creatives).
That's not the case with AI models. They can be infinitely replicated AND train on the output of all creatives. A comparable situation isn't one human learning from another human, it's millions of humans learning from every human. Only those humans don't even have to get paid, all their payment is funneled upwards.
It's not one artist vs. another artist, it's one artist against an army of infinitely replicable artists.
So this essentially boils down to an efficiency argument, and honestly it doesn't really address the core issue of whether it's 'stealing' or not.
What kind of creative types exist outside of living organisms? People can create award winning novels, but a table do not. Water do not. A paper with some math do not.
What is the basis that an LLM should be included as a "creative type"?
Well a creative type can be defined as an entity that takes other people's work, recombines it and then hides their sources.
LLMs seem to match.
Precisely. Nothing is truly original. To talk as though there's an abstract ownership over even an observation of the thing that force people to pay rent to use.. well artists definitely don't pay to whoever invented perspective drawings, programmers don't pay the programming language's creator. People don't pay newton and his descendants for making something that makes use of gravity. Copyright has always been counterproductive in many ways.
To go into details though, under copyright law there's a clause for "fair use" under a "transformative" criteria. This allows things like satire, reaction videos to exist. So long as you don't replicate 1-to-1 in product and purpose IMO it's qualifies as tasteful use.
What the fuck? People also need to pay to access that creative work if the rights owner charges for it, and they are also committing an illegal act if they don't. The LLM makers are doing this illegal act billions of times over for something approximating all creative work in existence. I'm not arguing that creative's make things in a vacuum, this is completely besides the point.
Never heard anything about what you are talking about. There isn't a charge for using tropes, plot points, character designs, etc. from other people's works if they are sufficently changed.
If an LLM reads a free wikipedia article on Aladdin and adds a genie to it's story, what copyright law do you think has been broken?
Meta and Anthropic atleast fed the entire copyrighted books into the training. Not the wikipedia page, not a plot summary or some tropes, they fed the entire original book into training. They used atleast the entirety of LibGen which is a pirated dataset of books.