I grew the apple, I should be able to decide what people do with it. I have a sign that says the apples are only for eating, but people are ignoring it.
For example, I don't read Medium articles as often as I used to, in part because I can just ask my programming questions directly to ChatGPT. Granted, this is to be expected because ChatGPT can customize it's response based on my unique situation whereas the Medium article is fixed, but it's still unfair to the author of the Medium article that ChatGPT was trained on, because without their effort, none of this would have been possible. And yet, because I am no longer generating as revenue for Medium, the original authors are no longer being compensated for the value they provided me.
I personally believe that LLM hosts should be required to pay for training data. With RAG, I think it makes sense to charge per-query, but for base models, content authors would probably need to organize into larger groups to facilitate payment.
Thereby increasing the amount of apples available to the population, lowering prices, making them available to more people. A net gain for society as a whole.
This happen all the time, you benefit from it all the time. You only have a problem with it because for once this system is not working in your direct favour.
Perhaps at first. A race to the bottom usually results in everyone being worse-off though — see red delicious apples.
For instance depending on jurisdiction it might be illegal to sell seeds that can make plants that can produce viable seeds, unless you have a specific authorisation and depending on the species.
Also, you know that morality is relative, and laws basically represent polities' morals, right ?
I sold an apple with DRM. One person bought my apple and gave it to a Chinese hacker, whom stripped the DRM and duplicated my apple for everyone, for free. They said it was not theft.
I grew an apple and put it on a cloud provider's server. I sold access to many duplicated apples, until a technology update forced me to change providers. I found out I could not take back my apples from the cloud providet without paying exorbitant fees. They said it was not theft.
I made an encrypted apple, I sold it on an app store. The app store banned me and my apple, the virus software flagged my apple, the TPM chip stopped my apple from being played, the government cracked open the encryption and duplicated my apple for national security purposes. They said it was not theft.
The only way to sell apples, is through perfectly-black, 'black boxes'. Putting apples on computers, gets them stolen. End of story.
It's been forty years now. There's nothing to be 'unsure about'.
'Benefit of the doubt' for intellectual property theft is deader than a dodo.
Edit: is “sufficiently transformative use” the metaphor? Is an LLM an apple pie baked out of the internet’s apple?
LLMs are advertised as 'language models', as if researchers coded in the Torah and the syntax of language of Eden in to a machine, and then it revealed all the truths of the world just by talking to itself. Just like AlphaZero did with chess. But in fact they are not 'language models'. They are models of what _people_ are saying. And people are, in fact, individual writers each providing their opinion and expertise. So it is a model of what Pete, Jack and Sarah said on the subject.
Except, Pete did it for add revenue, Jack to attract customers and Sarah for reputation. They won't get any of it, so they will stop writing. And so will end LLM's knowledge of novel subjects, because all it's knowledge is stolen IP. Thus, IP law is needed to protect actual producers of value.
To me, the problem with ChatGPT is that someone builds a business around my work, and I don't get paid or credited for it.
I feel the same way about content farms paraphrasing my work with no addition of their own. I'd also take offense to someone paraphrasing my website to paying customers as a service.