Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Judge dismisses DMCA copyright claim in GitHub Copilot suit

theregister.com

Post reply on HN
Print view View original

461–470 of 505 posts

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#461

Post by everforward »

cdrini wrote: post #408
sensanaty wrote: post #393

Earlier quoted context omitted.

They claim to not use anyone's private code, but I wouldn't trust the psychopathic C-suite at M$ not to murder kittens and human babies if it made the line go up a quarter of a percentage point, yet alone something like this.

You're free to speculate, but they have on multiple occasions said they don't train on private repos. Furthermore, there's no real incentive for them to do so, since (1) there are a lot of public repos, and (2) training on private repos opens them up to leaking things like private keys which would be a nightmare. It just doesn't make a lot of sense for them to do it.

Also, if anyone else uses private repos like I do, much of it will be "shame corner" code where it's only private because it's either half-finished or just terrible code because I wanted it to work right now.

My maintainable code gets published, my nightmares get banished to private repos so no one else thinks it's a good idea to replicate.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#462

Post by stale2002 »

account42 wrote: post #446
stale2002 wrote: post #313

Earlier quoted context omitted.

> acting like that's the immutable status quo It is immutable. What are you going to do about it? Confiscate everyone's home gamer PCs? Even in the most extreme hypothetical where lawsuits shutdown OpenAI, that doesn't delete the stable diffusion models that I have on my external hard drives. The tech is out there. It's too late.

Somehow this argument does not seem to hold for copyright enforcement of works that have been shared over BitTorrent and it's predecessors for decades.

I can start downloading any major/popular piece of media, starting right now, in under 60 seconds through bittorrent.

I cannot think of a better example of how futile copyright enforcement has been than the example that you just brought up.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#463

Post by everforward »

cellis wrote: post #198

I would like to ask an obvious question to the legally inclined here. How is this any different than remixing a song (lyrics/audio)? It's not "identical", and doesn't output "verbatim" lyrics or audio. What is the distinction between and . By a quick Google search it seems remixes violate copyright.

Not a lawyer, but that would be a fair use question, which I hear are notoriously complicated.

Colloquially, I generally expect a remix to be comprised of the original instrumentals/beat (potentially edited, but virtually nothing actually new added), potentially new lyrics, and to still be recognizable as the original.

The "still be recognizable as the original" part is a huge problem for fair use, and why I don't think remixes generally qualify. If it doesn't sound like the original then it's not a remix, but if it does sound like the original it can't be fair use.

I think the underlying issue is the resulting work, not the process that went into creating it. I think (but am in no way sure) that copying parts of songs would be fine if you did something to them so they aren't recognizable as the original.

As an example, if I take a song by the Beatles and repeatedly compress it until it's entirely compression artifacts, I would bet that I could publish that. I don't think it would matter that I started with a copyrighted work, what matters is that my finished product bears no resemblance to any other copyrighted work.

That would mean it's just a normal "is this work too similar to existing works?" standard applied to humans as well.

There is still an ancillary question of whether it's okay to train on copyrighted music, but that's really a different question than whether the works it creates infringe.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#464

Post by bryanrasmussen »

seanhunter wrote: post #407
bryanrasmussen wrote: post #342

Earlier quoted context omitted.

if that is the case why do people ever license covers? to clarify - I thought you just had to negotiate with the cover artist about rights and pay a nominal fee for usage of the song for cover purposes - that is to say you do not negotiate with the original artist, you negotiate with a cover artist and the whole process is cheaper?

You're maybe thinking about this in a way that's not helping you to understand the system and why it works the way it does. It's very clear when you think of a specific case. Say you want to make a recording of "Valerie" by the Zutons. You need permission (a license) from the songwriters (the Zutons presumably) to do this. You usually get this permission by paying a fee. Having done that, you can do your recording. W…

OK it seems exactly what I thought and described, and the opposite of what the parent poster described. The parent poster said that if you want to use the cover of the song you need to negotiate with both the people who did the cover and the original rights owner.

The closest I could get to a situation like that would be if I told Band B do a cover of Song A for my movie and I paid the licensing costs as part of my deal with Band B, but still not the same as the parent poster's description.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#465

Post by Kim_Bruning »

pclmulqdq wrote: post #434
Kim_Bruning wrote: post #308

Earlier quoted context omitted.

I actually once tracked this claim down in the case of stable diffusion. I concluded that it was just completely impossible for a properly trained stable diffusion model to reproduce the works it was trained on. The SD model easily fits on a typical USB stick, and comfortably in the memory of a modern consumer GPU. The training corpus for SD is a pretty large chunk of image data on the internet. That absolutely does…

I think this argument starts to break down for the (gigantic) GPTs where the model size is a lot closer to the size of the training corpus. Thinking in terms of compression, the compression in generative AI models is lossy. The mathematical bounds on compression only apply to lossless compression. Keeping in mind that a small fraction of the training corpus is presented to the training algorithm multiple times, it's…

Well, certainly up to GPT-3 that would seem a little odd. Models of somewhat similar capability are not THAT big, really. Eg:

  $ ollama list                            
  NAME                    ID              SIZE    MODIFIED     
  yi:34b                  ff94bc7c1b7a    19 GB   7 days ago  
  mistral:latest          61e88e884507    4.1 GB  2 months ago
  mixtral:8x22b           bf88270436ed    79 GB   2 months ago
  llama3:70b              be39eb53a197    39 GB   2 months ago
  phi3:latest             a2c89ceaed85    2.3 GB  2 months ago
  dolphin-mistral:latest  5dc8c5a2be65    4.1 GB  2 months ago
  yarn-mistral:7b-128k    6511b83c33d5    4.1 GB  2 months ago
  yarn-mistral:latest     8e9c368a0ae4    4.1 GB  2 months ago
  llama3:latest           a6990ed6be41    4.7 GB  2 months ago
For comparison, here's some stable diffusion checkpoints.

  ComfyUI/models/checkpoints $ du -h .
  6.5G    breakdomainxl_v03d.safetensors
  6.5G    dreamshaperXL10_alpha2Xl10.safetensors 
  6.5G    sd_xl_base_1.0.safetensors
  5.7G    sd_xl_refiner_1.0.safetensors
  ...
And I seem to recall there are some theoretical lower bounds on even lossy compression. Some quick back of the envelope fermi estimation gets me a hard lower bound of 5TB for "all the images on the internet"; but I'm not quite confident enough in my math to quite back that up right here and now.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#466

Post by loceng »

CodeWriter23 wrote: post #410
loceng wrote: post #394

Earlier quoted context omitted.

You just proved my point with your second sentence - that everything in the future will come. And bringing things more expediently is the actual opinion here, unsupported, where arguably it actually slows down not only progress but the value of that progress not being as widely distributed as it otherwise would be.

You continue to miss my point. Your point is a lazy, "the future will get here whenever it does" perspective. Mine is incentivizing discovery brings future innovations sooner.

You don't provide any supportive evidence that "your way of incentivizing" creates a net benefit - "just trust me bro" seems to be your fallback.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#467

Post by lumb63 »

xinayder wrote: post #317
lumb63 wrote: post #314

It seems to me that regardless of the outcome of this case, some developers do not want to have their code used to train LLMs. There may need to be a new license created to restrict this usage of software. Or, maybe developers will simply stop contributing open source. In today’s day and age, where open source code serves as a tool to pad Microsoft’s pockets, I certainly will not publish any of my software open sourc…

> I certainly will not publish any of my software open source, despite how much I would like to (under GPL) in order to help fellow developers. I think this is a rather radical approach. You're undermining the OSS movement because you dislike Microsoft (I do too). I think adding a clause or dual licensing your work is more effective at stopping big-tech funded AI crawlers than just not adhering to open source. You ca…

There is, as far as I am aware, nothing to stop Microsoft from crawling any other site’s code. Please correct me if I’m wrong. Things like “copyright” didn’t seem to stop OpenAI.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#468

Post by justinclift »

viridian wrote: post #460
justinclift wrote: post #270

Earlier quoted context omitted.

> Violate one or two copyrights, get sued or DMCAed out of existence. Violate billions, on the other hand, and you magically become immune to the rules everyone else has to follow. Sounds like the same concept as commonly said of "murderer vs conqueror". Could probably be applied to many other fields for disruption too. Not the murderer bit (!), more the "break one or two laws -> scaled up massively to a potential ne…

"If you owe the bank $100 that's your problem. If you owe the bank $100 million, that's the bank's problem."

Pretty sure there's a bunch of pre-existing laws around that though, so not really ripe for disrupting by scaling up the problem. ;)

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#469

Post by pclmulqdq »

Kim_Bruning wrote: post #465
pclmulqdq wrote: post #434

Earlier quoted context omitted.

I think this argument starts to break down for the (gigantic) GPTs where the model size is a lot closer to the size of the training corpus. Thinking in terms of compression, the compression in generative AI models is lossy. The mathematical bounds on compression only apply to lossless compression. Keeping in mind that a small fraction of the training corpus is presented to the training algorithm multiple times, it's…

Well, certainly up to GPT-3 that would seem a little odd. Models of somewhat similar capability are not THAT big, really. Eg: $ ollama list NAME ID SIZE MODIFIED yi:34b ff94bc7c1b7a 19 GB 7 days ago mistral:latest 61e88e884507 4.1 GB 2 months ago mixtral:8x22b bf88270436ed 79 GB 2 months ago llama3:70b be39eb53a197 39 GB 2 months ago phi3:latest a2c89ceaed85 2.3 GB 2 months ago dolphin-mistral:latest 5dc8c5a2be65 4.1…

> And I seem to recall there are some theoretical lower bounds on even lossy compression.

I'm not sure what your math is coming from and it seems trivially wrong. A single black pixel is a very lossy compression of every image on the internet. A picture of the Facebook logo is a slightly-less-lossy compression of every picture on the internet (the Facebook logo shows up on a lot of websites). I would believe that you can get a bound on lossy compression of a given quality (whatever quality means) only if you assume that there is some balance of the images in the compressed representation. There are a lot of assumptions there, and we know for a fact that the text fed to the GPTs to train them was presented in an unbalanced way.

In fact, if you look at the paper "textbooks are all you need" (https://arxiv.org/pdf/2306.11644) you can see that presenting a very limited set of information to an LLM gets a decent result. The remaining 6 trillion tokens in the training set are sort of icing on the cake.

Re: Judge dismisses DMCA copyright claim in GitHub Copilot suit

#470

Post by Kim_Bruning »

pclmulqdq wrote: post #469
Kim_Bruning wrote: post #465

Earlier quoted context omitted.

Well, certainly up to GPT-3 that would seem a little odd. Models of somewhat similar capability are not THAT big, really. Eg: $ ollama list NAME ID SIZE MODIFIED yi:34b ff94bc7c1b7a 19 GB 7 days ago mistral:latest 61e88e884507 4.1 GB 2 months ago mixtral:8x22b bf88270436ed 79 GB 2 months ago llama3:70b be39eb53a197 39 GB 2 months ago phi3:latest a2c89ceaed85 2.3 GB 2 months ago dolphin-mistral:latest 5dc8c5a2be65 4.1…

> And I seem to recall there are some theoretical lower bounds on even lossy compression. I'm not sure what your math is coming from and it seems trivially wrong. A single black pixel is a very lossy compression of every image on the internet. A picture of the Facebook logo is a slightly-less-lossy compression of every picture on the internet (the Facebook logo shows up on a lot of websites). I would believe that you…

Ok, that's a really low lower bound.

I think you'll agree that it would be a bit absurd to threaten legal action against someone for storing a single black pixel.

OTOH Someone might be tempted to start a lawsuit if they believe their image is somehow actually stored in a particular data file.

For this to be a viable class action lawsuit to pursue, I think you'd have to subscribe to the belief that it's a form of compression where if you store n images, you're also able to get n images back. Else very few people would have actual standing to sue.

Post reply on HN