3 ms·

Maybe I'm not aware, but as Meta's Llama is free for use, why is the lawsuit against them, vs OpenAI or any other company who have assuredly (from my experience

by aprilthird2021 2y ago

Maybe I'm not aware, but as Meta's Llama is free for use, why is the lawsuit against them, vs OpenAI or any other company who have assuredly (from my experience in these companies) done the exact same thing if not worse and make a profit from it?

consumer451 2y ago
I am far from a lawyer, but while it's free to use, Meta also uses Llama in their for-profit products. That is significant, is it not?
lxgr 2y ago
"Free to use" has never been legitimate excuse in copyright litigation. While commercial motives do matter (as far as I understand), being able to find evidence and bring a case practically matters even more.
- aprilthird2021 2y ago
  Yeah I understand that, for example, pirating copyrighted works for free is still illegal and can amount to damages for the copyright holder. Idk, just rubs me the wrong way when there are companies making money on the exact same product sourced the exact same way right now, as Meta chooses to make it free and gets sued. Seems like we should be logical enough to conclude if they get sued, every similar company should be investigated and fined (if wrongdoing was found) as well.
  - lxgr 2y ago
    Meta makes their models free to most users (not all – there's a "large commercial use" exemption!), and I do appreciate that practically. Still, they don't publish their detailed training methodology, which must have immense value for them internally. Even if they choose to never exercise the "large user exemption" in their current license, they can decide to license Llama 4 under restrictive terms (or not release weights at all – better start using Meta products if you want to gain access!) whenever it's convenient for them to do so. All of that doesn't exactly scream "public good worthy of a copyright exemption" to me, in a world where libraries, retro computing/gaming archivists and others are still continuously harassed by copyright holders.
patrickhogan1 2y ago
Evidence
- aprilthird2021 2y ago
  There is definitely lots of evidence in these companies internally that they ignored the fact that adding copyrighted works into the model might be legally grey. I hope someone steps up and whistle blows or shares the evidence from all the major AI companies. I think the most recent OpenAI whistleblower died before he could testify
Havoc 2y ago
They made the "mistake" of being honest in their initial papers about what the dataset used is. Everyone, including them, quickly changed their tune though. Now dataset is confidential trade secrets lol
est31 2y ago
I suppose it's easier to reach a deal with a company that is selling/renting out their model instead of giving it away for free, so they went against them first? Both from a budget at FB point of view, because probably they allocate a smaller budget to training than OpenAI does, so they can't be as generous, as it's not a core part of their business. And probably also from a point of view of the publishers, they probably don't like it either: with OpenAI they can do limited time deals, but with Llama the license allows redistribution so once it's published and a lot of business activity has been established on top of it, one can't come back and renegotiate, after say 10 years. It's in the best interest of big copyright to not have open(-ish) models in the ecosystem, they want entities they can seek rents from.
dialup_sounds 2y ago
0. There already are similar lawsuits against OpenAI https://www.documentcloud.org/documents/23963237-authors-v-openai/?responsive=1&title=1 https://www.documentcloud.org/documents/23963237-authors-v-o... 1. Because they literally said they used Books3 in the original Llama paper. The provenance of datasets used by other models is not as well documented. Books3 is known to be pirated. 2. Being free to use doesn't mitigate the authors' complaint in any way. (Compare: "I stole your bike, but then I gave it away.") The authors and artists (in the case of image models) want either a) to not have their work included in training sets or b) to be paid for that use via licensing. In either case they must enjoin the trainer of the model.
- aprilthird2021 2y ago
  Your points make sense. Thanks. For point 1, there are employees at OpenAI who do know the provenance of the datasets used and I am sure (based on my experiences) that it includes knowingly downloaded and inserted copyrighted works. Is not one single employee of OpenAI willing to blow the whistle?
  - dialup_sounds 2y ago
    Honestly, I doubt anybody cares that much. It's pretty much an open secret predicated on the untested idea that training is fair use, and the stakes aren't really that high in the long run even if it's not. Losses for the tech companies will just mean training data gets more expensive. They're all spending tens of billions on new data centers, so there's not even a question of whether they can afford it.
    - aprilthird2021 2y ago
      Yes, that's all true. Plus these days the datasets are sanitized from copyrighted works

consumer451 2y ago

I am far from a lawyer, but while it's free to use, Meta also uses Llama in their for-profit products. That is significant, is it not?

lxgr 2y ago

"Free to use" has never been legitimate excuse in copyright litigation. While commercial motives do matter (as far as I understand), being able to find evidence and bring a case practically matters even more.

aprilthird2021 2y ago
Yeah I understand that, for example, pirating copyrighted works for free is still illegal and can amount to damages for the copyright holder. Idk, just rubs me the wrong way when there are companies making money on the exact same product sourced the exact same way right now, as Meta chooses to make it free and gets sued. Seems like we should be logical enough to conclude if they get sued, every similar company should be investigated and fined (if wrongdoing was found) as well.
- lxgr 2y ago
  Meta makes their models free to most users (not all – there's a "large commercial use" exemption!), and I do appreciate that practically. Still, they don't publish their detailed training methodology, which must have immense value for them internally. Even if they choose to never exercise the "large user exemption" in their current license, they can decide to license Llama 4 under restrictive terms (or not release weights at all – better start using Meta products if you want to gain access!) whenever it's convenient for them to do so. All of that doesn't exactly scream "public good worthy of a copyright exemption" to me, in a world where libraries, retro computing/gaming archivists and others are still continuously harassed by copyright holders.

patrickhogan1 2y ago

Evidence

aprilthird2021 2y ago
There is definitely lots of evidence in these companies internally that they ignored the fact that adding copyrighted works into the model might be legally grey. I hope someone steps up and whistle blows or shares the evidence from all the major AI companies. I think the most recent OpenAI whistleblower died before he could testify

Havoc 2y ago

They made the "mistake" of being honest in their initial papers about what the dataset used is. Everyone, including them, quickly changed their tune though. Now dataset is confidential trade secrets lol

est31 2y ago

I suppose it's easier to reach a deal with a company that is selling/renting out their model instead of giving it away for free, so they went against them first? Both from a budget at FB point of view, because probably they allocate a smaller budget to training than OpenAI does, so they can't be as generous, as it's not a core part of their business. And probably also from a point of view of the publishers, they probably don't like it either: with OpenAI they can do limited time deals, but with Llama the license allows redistribution so once it's published and a lot of business activity has been established on top of it, one can't come back and renegotiate, after say 10 years. It's in the best interest of big copyright to not have open(-ish) models in the ecosystem, they want entities they can seek rents from.

dialup_sounds 2y ago

0. There already are similar lawsuits against OpenAI https://www.documentcloud.org/documents/23963237-authors-v-openai/?responsive=1&title=1 https://www.documentcloud.org/documents/23963237-authors-v-o... 1. Because they literally said they used Books3 in the original Llama paper. The provenance of datasets used by other models is not as well documented. Books3 is known to be pirated. 2. Being free to use doesn't mitigate the authors' complaint in any way. (Compare: "I stole your bike, but then I gave it away.") The authors and artists (in the case of image models) want either a) to not have their work included in training sets or b) to be paid for that use via licensing. In either case they must enjoin the trainer of the model.

aprilthird2021 2y ago
Your points make sense. Thanks. For point 1, there are employees at OpenAI who do know the provenance of the datasets used and I am sure (based on my experiences) that it includes knowingly downloaded and inserted copyrighted works. Is not one single employee of OpenAI willing to blow the whistle?
- dialup_sounds 2y ago
  Honestly, I doubt anybody cares that much. It's pretty much an open secret predicated on the untested idea that training is fair use, and the stakes aren't really that high in the long run even if it's not. Losses for the tech companies will just mean training data gets more expensive. They're all spending tens of billions on new data centers, so there's not even a question of whether they can afford it.
  - aprilthird2021 2y ago
    Yes, that's all true. Plus these days the datasets are sanitized from copyrighted works