6 ms·

It's unfortunate we are seeing all of these data platforms get locked off, because this is not going to affect AI development from big companies, it's only goin

It's unfortunate we are seeing all of these data platforms get locked off, because this is not going to affect AI development from big companies, it's only going to affect the ability for individuals to run AI development of any form in their home. I hope the data that has been found so far is going to big enough going forward, but it's incredibly unfortunate that this is happening. I hope all the people making these decisions wake up with a bad headache and severe heartburn tomorrow.

CoastalCoder 3y ago
IANAL, but I'm curious: Suppose that deep-pocketed AI companies were paying Reddit, Stack Overflow, etc. to make it harder for other AI upstarts to access those data. I.e., to build a mote by denying competitors access to previously accessible data sets. Would that violate antitrust laws in various major markets?
- bioemerl 3y ago
  Given that this seems to happen all the time without antitrust issues it probably wouldn't, even though I feel like it should. What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it.
  - josephcsible 3y ago
    > What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it. Couldn't that be accomplished by a law or ruling that using something for training AI doesn't exempt you from having to follow its license? OpenAI is already in blatant violation of both the "BY" and "SA" parts of the existing license.
    - juliangoldsmith 3y ago
      Arguably, a model created by training on a corpus of data is a derived work of that corpus. Let's say I take a collection of images and use a program to compress them. When decompressed, the images are close to, but not exactly the same as the originals. Despite being in a different format, and despite not being exactly the same as the originals, the copyright to the compressed images is still held by whoever previously held it. If I take the collection of images from earlier and train a diffusion model based on it, I'm essentially just compressing it a different way. With the right prompt, you can get out something very similar to what you put in.
      - rileymat2 3y ago
        By this logic, isn't remembering something in your brain also a derived work? But that would not make any sense to protect until you create and distribute something based on that memory. The same logic should be applied to this.
        jacksnipe 3y ago
        If you remember it from your brain and perform it live, that’s perfectly fine. So there’s a line to be drawn somewhere and I don’t think it’s super clear cut in most cases.
        bioemerl 3y ago
        No, according to copyright if you remember something from your brain and perform it live you are very much in violation. If you remember it and make something that is a distinct work, something that may be steals the idea without reproducing any of its elements, that's never been considered under copyright. I think that's going to be the litmus test for these AI. If you can get them to produce out both that is this things from anything else, it's not going to be a copyright violation because it's not a copy of anything.
      - bioemerl 3y ago
        > If I take the collection of images from earlier and train a diffusion model based on it, I'm essentially just compressing it a different way Not really. If diffusion models were compression they'd be so lossy as to be totally worthless
  - endisneigh 3y ago
    > What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it. inherently not possible as then it would not be "open" to begin with.
    - bioemerl 3y ago
      Open except you have to pay if it gets big enough seems perfectly reasonable to me. I understand the idea, it's not truly open in that case, but so long as the ability to build new things on it and prosper from it is preserved im alright. The key is that it's not doing something like trying to restrict you from using it in a certain way, only requiring you give a fair share of profits. This was, fun fact, the original purpose of patents. They weren't designed to keep things closed and owned by individuals, they were designed to allow people to freely share and make a profit so that ideas could be built on by each other. The patent system is turned into this corrupted terrible mess where things are almost never shared through licensing or payment, and it's just a way to build monopolistic enterprises nowadays. An open source system that allows for this sort of payment would also allow for many many more things to be open when currently the bad actors who will build and take that work and just never pay you back for it.
    - josephcsible 3y ago
      There are ways to require payment for some uses of things that are legitimately open. As an example, consider the practice of selling exceptions to the GPL, as is done for Qt.
      - endisneigh 3y ago
        > As an example, consider the practice of selling exceptions to the GPL, as is done for Qt. There are people who do not consider that "open". See the whole debacle about what exactly constitutes "open source"
        josephcsible 3y ago
        > There are people who do not consider that "open". Who? Even Richard Stallman is okay with what they do: https://www.fsf.org/blogs/rms/selling-exceptions https://www.fsf.org/blogs/rms/selling-exceptions
- semi-extrinsic 3y ago
  Nitpick, also because the contrast is kind of funny: mote: a small particle, speck, atom, "mote of dust" moat: a deep ditch, often filled with water, as a first line of defence around a castle.
  - btown 3y ago
    Hopefully this comment won't be demoated by the algorithm - it truly holds water on its own!
    - sitkack 3y ago
      You are only helping it with puns and we know that puns are the gateway to consciousness! They are like the corpus callosum of language, serving as the bridge that spans the moat between the castles of wit and creativity.
grumbel 3y ago
> It's unfortunate we are seeing all of these data platforms get locked off Are there any AGPL-like licenses that address this?

CoastalCoder 3y ago

IANAL, but I'm curious: Suppose that deep-pocketed AI companies were paying Reddit, Stack Overflow, etc. to make it harder for other AI upstarts to access those data. I.e., to build a mote by denying competitors access to previously accessible data sets. Would that violate antitrust laws in various major markets?

bioemerl 3y ago
Given that this seems to happen all the time without antitrust issues it probably wouldn't, even though I feel like it should. What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it.
- josephcsible 3y ago
  > What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it. Couldn't that be accomplished by a law or ruling that using something for training AI doesn't exempt you from having to follow its license? OpenAI is already in blatant violation of both the "BY" and "SA" parts of the existing license.
  - juliangoldsmith 3y ago
    Arguably, a model created by training on a corpus of data is a derived work of that corpus. Let's say I take a collection of images and use a program to compress them. When decompressed, the images are close to, but not exactly the same as the originals. Despite being in a different format, and despite not being exactly the same as the originals, the copyright to the compressed images is still held by whoever previously held it. If I take the collection of images from earlier and train a diffusion model based on it, I'm essentially just compressing it a different way. With the right prompt, you can get out something very similar to what you put in.
    - rileymat2 3y ago
      By this logic, isn't remembering something in your brain also a derived work? But that would not make any sense to protect until you create and distribute something based on that memory. The same logic should be applied to this.
      - jacksnipe 3y ago
        If you remember it from your brain and perform it live, that’s perfectly fine. So there’s a line to be drawn somewhere and I don’t think it’s super clear cut in most cases.
        bioemerl 3y ago
        No, according to copyright if you remember something from your brain and perform it live you are very much in violation. If you remember it and make something that is a distinct work, something that may be steals the idea without reproducing any of its elements, that's never been considered under copyright. I think that's going to be the litmus test for these AI. If you can get them to produce out both that is this things from anything else, it's not going to be a copyright violation because it's not a copy of anything.
    - bioemerl 3y ago
      > If I take the collection of images from earlier and train a diffusion model based on it, I'm essentially just compressing it a different way Not really. If diffusion models were compression they'd be so lossy as to be totally worthless
- endisneigh 3y ago
  > What we need is a legal way for companies to keep the data open, but also require OpenAI and friends to pay them for it. inherently not possible as then it would not be "open" to begin with.
  - bioemerl 3y ago
    Open except you have to pay if it gets big enough seems perfectly reasonable to me. I understand the idea, it's not truly open in that case, but so long as the ability to build new things on it and prosper from it is preserved im alright. The key is that it's not doing something like trying to restrict you from using it in a certain way, only requiring you give a fair share of profits. This was, fun fact, the original purpose of patents. They weren't designed to keep things closed and owned by individuals, they were designed to allow people to freely share and make a profit so that ideas could be built on by each other. The patent system is turned into this corrupted terrible mess where things are almost never shared through licensing or payment, and it's just a way to build monopolistic enterprises nowadays. An open source system that allows for this sort of payment would also allow for many many more things to be open when currently the bad actors who will build and take that work and just never pay you back for it.
  - josephcsible 3y ago
    There are ways to require payment for some uses of things that are legitimately open. As an example, consider the practice of selling exceptions to the GPL, as is done for Qt.
    - endisneigh 3y ago
      > As an example, consider the practice of selling exceptions to the GPL, as is done for Qt. There are people who do not consider that "open". See the whole debacle about what exactly constitutes "open source"
      - josephcsible 3y ago
        > There are people who do not consider that "open". Who? Even Richard Stallman is okay with what they do: https://www.fsf.org/blogs/rms/selling-exceptions https://www.fsf.org/blogs/rms/selling-exceptions
semi-extrinsic 3y ago
Nitpick, also because the contrast is kind of funny: mote: a small particle, speck, atom, "mote of dust" moat: a deep ditch, often filled with water, as a first line of defence around a castle.
- btown 3y ago
  Hopefully this comment won't be demoated by the algorithm - it truly holds water on its own!
  - sitkack 3y ago
    You are only helping it with puns and we know that puns are the gateway to consciousness! They are like the corpus callosum of language, serving as the bridge that spans the moat between the castles of wit and creativity.

grumbel 3y ago

> It's unfortunate we are seeing all of these data platforms get locked off Are there any AGPL-like licenses that address this?