Yeah but we got all the dys without any of the topia. I was promised high quality prosthetics, neon blinkenlights, and the right to bear arms. We’ve got like 15% of the appropriate level of any of those.
Kind of a strawman, I’d like everything to be FOSS, and if we keep Capitalism (which we shouldn’t), it should be HEAVILY regulated not the laissez-faire corporatocracy / oligarchy we have now.
I don’t want any for-profit capitalists to have any control of AI. It should all be owned by the public and all productive gains from it taxed at 100%. But open source AI models, right on.
A website where you can download paywalled scientific literature. Most scientific literature is paywalled by publishers, and costs a real significant amount to read (like 30-50$ per article if you don’t have a subscription).
Scihub basically just pirates it. And has been shut down several times. But as most scientific studies are already laid with public money, scihub isn’t that unethical at all.
Lots of scientists will just send you their article if you email them. They don’t get the money when you pay to read it - often they pay to submit. Reviewing journal articles is a privilege and doesn’t get you paid. The prestige of a scientific article is from the number of times people have cited it. The only “harm” done is that the publisher doesn’t get to make 100% profit for doing nothing.
Journal publishing is mostly a way to extract money from universities. Elsevier and its ilk name whatever price they think a research university can afford.
Very true. Also, a new federal policy is now in place and requires any research funded even in part by federal money be open access. As a result we should see much more high quality research becoming open access (already has begun). Only downside is research labs like mine have to use more money to publish to these journals because open access costs more for the authors. Hopefully this system gets reformed during my lifetime.
But yes, please just email the authors! Works most of the time and I think it’s fun.
A.I. doesn’t violate copywrite laws. It is the data-mining done to train A.I. and the regurgitation of said data in the responses that ultimately violate these laws. A model trained on privately owned, properly licensed, or exclusively public works wouldn’t be a problem.
Even then, I would argue that lack of attribution is a bigger problem than merely violating copywrite. A big part of the LLM mystique is in how it can spit out a few lines of Shakespeare without accreditation and convince its users that its some kind of master poet.
Copywrite law is stupid and broken. But plagarism is a problem in its own right, as it seeks to effectively sell people their own creative commons at an absurd markup.
A model trained on privately owned, properly licensed, or exclusively public works wouldn’t be a problem.
This is how we end up with only corpo owned AIs being allowed to exist imo, places like stock photo sites are the only ones with large enough repositories of images to train AI that they have all the legal rights to
The way I see it, either generative AI is legal, free for everyone to run locally, and the created works are public domain, OR, everyone pays $20/mo to massive faceless corpos for the rest of their lives to have the privilege of access to it because they’re the only ones who own all (or have enough money to license) the IP needed to train them
This is how we end up with only corpo owned AIs being allowed to exist imo
Its how you end up with sixteen different streaming services that only vend a sliver of the total available content, sure. But the underlying technology of AI grows independent of what its trained on.
The way I see it, either generative AI is legal, free for everyone to run locally, and the created works are public domain, OR, everyone pays $20/mo to massive faceless corpos for the rest of their lives to have the privilege of access to it
There are other alternatives. These sites can be restricted to data within the public domain. And we can increase our investment in public media. The problem of NYT articles being digested and regurgitated as ChatGPT info-vomit isn’t a problem if the NYT is a publicly owned and operated enterprise. Then its not struggling to profit off journalism, but treating this information as a loss-leading public service open to all, with ChatGPT simply operating as a tool to store, process, and present the data.
Similarly, if you limit generative AI to the old Mickey Mouse and Winnie-the-Pooh films from the 1930s, you leave plenty of room for original artists to create new works without fear that their livelihoods get chews up and fed back into the system. If you invest in public art exhibitions then these artists can get paid to pursue their craft, the art becomes public domain immediately, and digital tools that want to riff on the original are free to do so without undermining the artists themselves.
Yes, because 1:1 duplication of copy written works violates copyright, but summaries of those works and relaying facts stated in those works is perfectly legal (by an ai or not).
If you mean by “perfectly legal” a fair use claim, then could you please explain how a commercial for-profit company using the works, sometimes echoing verbatim results, is infringing on the copyrights in a fair use manner?
I do not mean a fair use claim. To quote the copyright office “Copyright does not protect facts, ideas, systems, or methods of operation, although it may protect the way these things are expressed” source
Facts and ideas cannot be copy written, so what I was specifically referring to is that if I or an AI read a paper about jellyfish being ocean creatures, then later talk about jellyfish being ocean creatures, there’s no restrictions on that whatsoever as long as we don’t reproduce the paper word by word.
Now, most of the time AI summarizes things or collects facts, and since those themselves cannot be protected by copyright it’s perfectly legal. On the occasion when AI spits out copy written work then that’s a gray area and liability if any will probably decided in the courts.