cross-posted from: https://lemmy.intai.tech/post/43759
cross-posted from: https://lemmy.world/post/949452
OpenAI’s ChatGPT and Sam Altman are in massive trouble. OpenAI is getting sued in the US for illegally using content from the internet to train their LLM or large language models
i posted on the public internet with the intent and understanding that it would be crawled by systems for all kinds of things. if i dont want content to be grabbed i dont publish it publicly
you can’t easily have it both ways imo. even with systems that do strong pki if you want the world in general to see it you are giving up a certain amount of control over how the content gets used.
law does not really matter here as much as people would like to try to apply it, this is simply how public content will be used. Go post in a garden if you don’t want to get scrapped, just remember the corollary is your reach, your voice is limited to the walls of that garden.
What you said makes a lot of sense. But here’s the catch: it assumes OpenAI checked the licensing for all the stuff they grabbed. And I can guarantee you they didn’t.
It’s damn near impossible to automatically check the licensing for all the stuff they got she we know for a fact they got stuff whose licensing does not allow it to be used this way. Microsoft has already been sued for Copilot, and these lawsuits will keep coming. Assuming they somehow managed to only grab legit material and they used excellent legal advisors that assured them out would stand in court, it’s definitely impossible to tell what piece of what goes where after it becomes a LLM token, and also impossible to tell what future lawsuits will decide about it.
Where does that leave OpenAI? With the good ol’ “I grabbed something off the internet because I could”. Why does that sound familiar? It’s something people have been doing since the internet was invented, it’s commonly referred to as “piracy”. But it’s supposed to be wrong and illegal. Well either it’s wrong and illegal for everybody or the other way around.
there were court cases around this very thing and google and webarchive. I suspect thier legal team is expecting similar precedent with the issue being down to the individual and how they use the index, example, using it to make my own unique character (easily done) vs making an easy and obvious rip off of a Disney property. The same tests can be applied, the question IMO isn’t about the index that is built here. I can memorize a lot (some people have actual eidetic memory) and synthesize it too which is protected and I can copyright my own mental outputs. The disposition of this type of output vs mechanical outputs i expect will be where things end up being argued.
I’m not going to say I’m 100% right here, we are in a strange timeline but there is precedent for what OAI is doing IMO.
The issue becomes the sale/profit of selling access, such as with GPT-4 right now. Indexing/archiving and selling are two very different beasts.
interesting lines to walk, depends on what they are selling, there is a definte cost to running a model and you are allowed to charge a reasonable fee to handle the process of providing the records. we used to pay per page for this kind of thing, now you pay per token
they can also sell a lot of services and tools around the model while still not using it in a non-infringing manner. this will all end up in front a of a judge, with the books laid out i suspect. I am not sure we will ever see any of the details, i hope we do.
The difference between piracy and having your content used for training a generative model, is that in the latter case, the content isn’t redistributed. It’s like downloading a movie from netflix (and eventually distributing it for free) vs watching a movie on netflix and using it as inspiration to make your own movie.
The legality of it all is unclear and most of that is because the technology evolved so quickly that the legal framework is just not equipped to deal with it. Despite the obvious moral issues with scraping artist’s content.
Yes, notice I said “Scrape” and not “steal” :)
abs, providing some more background from my perspective.