4 ms·

Publishing a derivative work of something you scraped is not permitted. The question is whether LLMs are derivative works, which has not been settled in courts

Publishing a derivative work of something you scraped is not permitted. The question is whether LLMs are derivative works, which has not been settled in courts yet.

Kim_Bruning 3y ago
Technically correct on the point that it has not been settled in courts yet. For eg stable diffusion I actually dove a bit deeper on this: The LAION dataset is -in itself- almost indisputably legal: It contains URLs of images, and tags describing those images[1]. To be sure: _It does not contain actual images_. People have made this mistake before, and LAION quite correctly reply that they don't have them. [2] Stable diffusion's models are much smaller [3] than the LAION dataset which they use as their source. Again, the models do not contain images. This is pretty suggestive that SD models should not be considered derivative works. Of course sometimes funny things happen in court, so we can't be sure until a court actually decides. I don't currently run my own GPT or LLAMA, but those should have similar properties. (in fact, OpenAI appears to claim so.) [1] https://laion.ai/blog/laion-5b/ https://laion.ai/blog/laion-5b/ (if you look for the files in any of the variations, you'll see it's in the order of terrabytes) [2] https://www.vice.com/en/article/pkapb7/a-photographer-tried-to-get-his-photos-removed-from-an-ai-dataset-he-got-an-invoice-instead https://www.vice.com/en/article/pkapb7/a-photographer-tried-... [3] https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 https://huggingface.co/stabilityai/stable-diffusion-xl-base-... : see under "files and versions", it's sd_xl_base_1.0.safetensors. File size is 6.9 GB
- Certhas 3y ago
  Not sure that the size of the model is relevant. If you take a PNG of Mickey Mouse, and create a similar vector graphic that is a fraction of the size, that would still be derivative.
- satyrnein 3y ago
  Thanks for the links! To me (merely an amateur observer), it is not perfectly obvious that the LAION dataset is legal, because I don't think "we don't have your original images anymore, we only kept the results of applying an algorithm to them" is a universal defense. If the algorithm is "resize to 640×480" you are probably still infringing, but if it's "calculate md5 checksum" you're probably not. I don't know where on the spectrum LLM training will fall, legally!
  - Kim_Bruning 3y ago
    Heh, it does in fact calculate the MD5 checksum! In total, the fields in the LAION dataset are: image_path caption NSFW similarity LICENSE url key status error_message width height original width original height exif md5 https://huggingface.co/datasets/laion/laion2b-multi-vit-l-14-embeddings https://huggingface.co/datasets/laion/laion2b-multi-vit-l-14... Note that LAION is a metadata dataset, it's not an image generation model itself. (But it has been used to train one)
fortyseven 3y ago
I do not trust a court to weigh in properly on the deep technical nuances involved.
- Guvante 3y ago
  These kinds of things are not settled by arbitrary judges making off the cuff decisions. They are settled by specialist technical judges with advice of dozens of specialists explaining the tech in detail. Check out Google vs Oracle the lawsuit about whether an API is copyrightable for some examples of how the rulings are very much based on deep understanding. The only downside is the process is slow and needs both a plaintiff and defendant willing to spend hundreds of thousands or millions of dollars to settle the case in court. It also takes many months of time.
  - nextaccountic 3y ago
    last I heard about this ruling was that APIs are copyrightable, which is terrible for open source reimplementations of all stuff
    - Guvante 3y ago
      More trying to focus on how the ruling wasn't naive as OP implies all rulings are by saying "I wouldn't trust a judge". Alternatively they could mean "judges should stay away from X" but that is nonsensical given the courts being arbitrage...
- nextaccountic 3y ago
  we're talking about a law (copyright), who else can weight in besides the courts? or are you suggesting abolishing copyright? because I'm all for it
  - janosdebugs 3y ago
    This is an interesting topic that comes up time and time again. I'd like to understand, if we abolish copyright and I make a computer game, pour all my money and time in it, what's to stop EvilBigPublisher Co from simply taking it, rebranding it and killing my company with their might? Or, if I came across a copy of Unreal Engine, what would stop me from using it and not paying Epic a dime?

Publishing a derivative work of something you scraped is not permitted. The question is whether LLMs are derivative works, which has not been settled in courts yet.

Kim_Bruning 3y ago

Technically correct on the point that it has not been settled in courts yet. For eg stable diffusion I actually dove a bit deeper on this: The LAION dataset is -in itself- almost indisputably legal: It contains URLs of images, and tags describing those images[1]. To be sure: _It does not contain actual images_. People have made this mistake before, and LAION quite correctly reply that they don't have them. [2] Stable diffusion's models are much smaller [3] than the LAION dataset which they use as their source. Again, the models do not contain images. This is pretty suggestive that SD models should not be considered derivative works. Of course sometimes funny things happen in court, so we can't be sure until a court actually decides. I don't currently run my own GPT or LLAMA, but those should have similar properties. (in fact, OpenAI appears to claim so.) [1] https://laion.ai/blog/laion-5b/ https://laion.ai/blog/laion-5b/ (if you look for the files in any of the variations, you'll see it's in the order of terrabytes) [2] https://www.vice.com/en/article/pkapb7/a-photographer-tried-to-get-his-photos-removed-from-an-ai-dataset-he-got-an-invoice-instead https://www.vice.com/en/article/pkapb7/a-photographer-tried-... [3] https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 https://huggingface.co/stabilityai/stable-diffusion-xl-base-... : see under "files and versions", it's sd_xl_base_1.0.safetensors. File size is 6.9 GB

Certhas 3y ago
Not sure that the size of the model is relevant. If you take a PNG of Mickey Mouse, and create a similar vector graphic that is a fraction of the size, that would still be derivative.
satyrnein 3y ago
Thanks for the links! To me (merely an amateur observer), it is not perfectly obvious that the LAION dataset is legal, because I don't think "we don't have your original images anymore, we only kept the results of applying an algorithm to them" is a universal defense. If the algorithm is "resize to 640×480" you are probably still infringing, but if it's "calculate md5 checksum" you're probably not. I don't know where on the spectrum LLM training will fall, legally!
- Kim_Bruning 3y ago
  Heh, it does in fact calculate the MD5 checksum! In total, the fields in the LAION dataset are: image_path caption NSFW similarity LICENSE url key status error_message width height original width original height exif md5 https://huggingface.co/datasets/laion/laion2b-multi-vit-l-14-embeddings https://huggingface.co/datasets/laion/laion2b-multi-vit-l-14... Note that LAION is a metadata dataset, it's not an image generation model itself. (But it has been used to train one)

fortyseven 3y ago

I do not trust a court to weigh in properly on the deep technical nuances involved.

Guvante 3y ago
These kinds of things are not settled by arbitrary judges making off the cuff decisions. They are settled by specialist technical judges with advice of dozens of specialists explaining the tech in detail. Check out Google vs Oracle the lawsuit about whether an API is copyrightable for some examples of how the rulings are very much based on deep understanding. The only downside is the process is slow and needs both a plaintiff and defendant willing to spend hundreds of thousands or millions of dollars to settle the case in court. It also takes many months of time.
- nextaccountic 3y ago
  last I heard about this ruling was that APIs are copyrightable, which is terrible for open source reimplementations of all stuff
  - Guvante 3y ago
    More trying to focus on how the ruling wasn't naive as OP implies all rulings are by saying "I wouldn't trust a judge". Alternatively they could mean "judges should stay away from X" but that is nonsensical given the courts being arbitrage...
nextaccountic 3y ago
we're talking about a law (copyright), who else can weight in besides the courts? or are you suggesting abolishing copyright? because I'm all for it
- janosdebugs 3y ago
  This is an interesting topic that comes up time and time again. I'd like to understand, if we abolish copyright and I make a computer game, pour all my money and time in it, what's to stop EvilBigPublisher Co from simply taking it, rebranding it and killing my company with their might? Or, if I came across a copy of Unreal Engine, what would stop me from using it and not paying Epic a dime?