this post was submitted on 23 Sep 2023
210 points (91.0% liked)
Technology
59322 readers
5220 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I don't think the issue is corps feeding the internet into AI systems. The real issue is gatekeeping to information and only giving access to this information while milking the individual for data by trackers, money by subscriptions, and more money by ads (that we pay for with subscriptions).
Another larger issue that I fear is often ignored is the amount of control large corporations and in theory the government can have over us just by looking at our trace we leave in the internet. Just have a look at Russia and China for real world examples of this.
As an open source contributor, I believe information (facts and techniques) should be free.
As an open source contributor, I also know that two-way collaboration only happens when users understand where the software came from and how they can communicate back to the original author(s).
The layer of obfuscation that LLMs add, where the code is really from XYZ open-source project, but appears to be manifesting from thin air... worries me, because it's going to alienate would-be collaborators from the original authors.
"AI" companies are not freeing information. They are colonizing it.
The code that AI produces isn't "copied" from those original authors, though. The AI learned how to code from them, it isn't literally copying and pasting from them.
If you think a bit of code is "really from" XYZ open-source project, that's a copyright violation and you can pursue that legally. But you'll need to actually show that the code is a copy.
Your justification seems to rest on whether LLM training technically passes the legal standard of violating IP.
That's not a super compelling argument to me, because:
I don't think that there is fundamentally anything wrong with LLMs as a technology. My problem is that the economic incentives are misaligned with long-term stability of the creative pools that fuel these things in the first place.
That's basically all that I'm talking about here, yeah. I'm saying that the current laws don't appear to say anything against training AIs off of public data. The AI model is not a copy of that data, nor is its output.
Indeed. Things are not illegal by default, there needs to be a law or some sort of precedent that makes them illegal. In the realm of LLMs that's very sparse right now for exactly the reason you say. Nobody anticipated it so nobody wrote any laws forbidding it.
There are things that you can use intellectual property for that do not require consent in the first place. Fair use describes various categories of that. If it's not illegal to use copyrighted material without permission when training AIs, why would it matter whether the license permitted it or the author consented to it?
Wouldn't requiring licensing of data for the training of LLMs stack things even more in the favour of big IP-owning platforms?
Again, as I said before, if you think some specific bit of LLM output is violating the copyright of some code you wrote, there's already laws in place specifically covering that situation. You can go to court and show that the two pieces of code are substantially identical and sue for damages or whatever. The AI model itself is another matter, though, and I doubt any current laws would count it as a "copy" of the data that went into training it.
The copyright violation has happened when the code got fed into that AI's greedy gullet, not when it came out of it's rear end.
That remains to be tested legally speaking, and I don't think it's likely to pass muster. If it was trained correctly (ie, no overfitting) the resulting AI model does not contain a copy of the training inputs in any identifiable sense.
Yes, the laws are probably muddy in Usa as usual, but rather clear here in the EU. But legal proceedings are slow, and Big Tech is making haste with their feeding.
There are many jurisdictions beyond the US and EU, Japan in particular has been very vocal about going all-in on allowing AI training. And I wouldn't say the EU's laws are "clear" until they are actually tested.
My open source project benefits hugely from the free to access LLM coding tools available, that's a far bigger positive than the abstract fear that someone might feel alienated because the guy copy pasting their code doesn't know who he's copying from?
And yes, obviously the LLM isn't copying code it's leaning from a huge range of sources and combining it to make exactly what you ask for (well not exactly but with some needling it gets there eventually) but even if it were that's still not disrupting collaboration because that's not how collaboration works - no one says 'instead of coding all the boring elif statements required for my fiction determining if something is a prime, I'll search code snippits and collaborate with them' every worthwhile collaborator to my project has been an active user of the software and wanted to help improve it or add functions - AI won't change that, and if it does it'll only be because it makes coding so easy I don't need collaborators
Yep, the truly free and open internet is coming to an end. Corporations and governments have spent decades trying to claim control over it, and they're nearly there.
Which, ironically, will be greatly expedited by the drive to prohibit AI from learning from "unlicensed" materials. That will guarantee that the only AIs with a broad training set will be those owned by corporations that already control an enormous amount of training materials (Disney, Getty Images, etc.)
Yeah, right now the fight is between corporations and creators, but I feel like the future battle is going to be between corporate AIs and "pirated" ones, because Disney is going to keep a firm chokehold over what its generative AI can make, while the community ones will completely ignore copyright restrictions and just let people do whatever they want.
Not gonna need to worry about paywalls when you can get a pirated generative AI to create the superhero mashup you always wanted to watch as a child. That said, I could definitely see Disney and other piggybacking off of AI panic to extend copyright protection into spaces that were previously fair use.
A factor I didn't consider. Thanks. And there I thought given hardware requirements it would be relatively easy to build such LLMs or similar foss-like.