Besides, you really don’t need to zoom in on “but muh license agreement” to roast these AI turds.
They’re very clear: We’re gonna put creatives out of work, we’re gonna sell a unified product to replace them, and we’re gonna use their own labor to build their replacements.
That’s anticompetitive.
Nail em on that instead of trying to thread the needle on reining in the tech lords without damaging e.g. linguistic analysis researchers.
We’re gonna put creatives out of work, we’re gonna sell a unified product to replace them, and we’re gonna use their own labor to build their replacements.
Yes, but: it’s short sighted, and wrong. Until we have a sea change in the LLM/AGI space, “creatives” will be needed for seed data. LLMs that are recursively trained on their own output degrade and produce worse output over time.
The “yes” part is that companies looking to replace paying people for their work, but still hoping that Creative Commons types are still posting online for free harvesting.
Yeah I’m just downloading random data for fun in little tiny bits. If that data happens to arrange itself in the form of the latest episode of Doctor Who then that’s not my problem.
I think it’s really disingenuous to mention the DeviantArt/Midjourney/Runway AI/Stability AI lawsuit without talking about how most of the infringement claims were dismissed by the judge.
Yes, that would also be statistical correlations to an AI model. The specific kind of information they’re being trained on doesn’t affect the underlying mechanism of model training.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !piracy@lemmy.dbzer0.com
⚓ Dedicated to the discussion of digital piracy, including ethical problems and legal advancements.
These files are just correlated bits and bytes, nothing more.
Aren’t MP3s just a statistical correlation?
Besides, you really don’t need to zoom in on “but muh license agreement” to roast these AI turds.
They’re very clear: We’re gonna put creatives out of work, we’re gonna sell a unified product to replace them, and we’re gonna use their own labor to build their replacements.
That’s anticompetitive.
Nail em on that instead of trying to thread the needle on reining in the tech lords without damaging e.g. linguistic analysis researchers.
The tools exist for creatives to use.
Yes, but: it’s short sighted, and wrong. Until we have a sea change in the LLM/AGI space, “creatives” will be needed for seed data. LLMs that are recursively trained on their own output degrade and produce worse output over time.
The “yes” part is that companies looking to replace paying people for their work, but still hoping that Creative Commons types are still posting online for free harvesting.
Why not go full data nihilist and say that every file is just a natural number expressed in binary.
I’m good with that
Yeah but I legally own this particular number >:E
Torrents are just statistical correlations to our torrent clients.
Siri create a random dataset in the likeness of peter jacksons lord of the rings: the return of the king directors cut on dvd
Copies are just very strong statistical correlations.
Yeah I’m just downloading random data for fun in little tiny bits. If that data happens to arrange itself in the form of the latest episode of Doctor Who then that’s not my problem.
Reminder for everyone who does not know: doctor who is available on the bbc iplayer, which can be downloaded with get-iplayer.
If you are outside the UK, all you need is a DNS or VPN.
Damn, this article is so biased.
Seemed pretty fair and fact-based to me. What bias are you seeing?
I think it’s really disingenuous to mention the DeviantArt/Midjourney/Runway AI/Stability AI lawsuit without talking about how most of the infringement claims were dismissed by the judge.
What about copyrighted code?
Like for instance, GPU drivers?
Yes, that would also be statistical correlations to an AI model. The specific kind of information they’re being trained on doesn’t affect the underlying mechanism of model training.
I mentioned it before:
If they use any GPL code for their model then any output would be a derived work and a violation of the GPL.