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

I do not agree with Github's use of copyrighted code as training for Copilot

thelig.ht

Post reply on HN
Print view View original

211–220 of 545 posts

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#211

Post by kennywinker »

owlbynight wrote: post #201

Has this person been in a coma? If I utilize a free service on the Internet, I'm trading for some kind of convenience with the knowledge that I am in some way being boned in the backend by teams of people, all of whom are likely more clever than I am and using my patronage to some kind of nefarious end. The Internet isn't really a place to exercise an inflexible moral code. His new repository probably can be traced b…

> His new repository probably can be traced back to slave labor somehow

And you're ok with that? It doesn't HAVE to be like this. Just because you've chosen nihilism, doesn't mean that's the only choice, and it certainly doesn't help anything.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#213

Post by kingsuper20 »

caconym_ wrote: post #131
henvic wrote: post #21

I really hope this weakens copyright. We can live without it.

Copyright doesn't just benefit huge corporations. For instance, without it, independent artists who rely on copying for distribution (authors, musicians, etc.) would find it much more difficult to make money off their work, mostly (IMO) because large corporate entities with large investments made in publication and distribution systems could simply take content and sell it themselves with zero obligation to the origi…

>Copyright doesn't just benefit huge corporations. For instance, without it, independent artists who rely on copying for distribution (authors, musicians, etc.) would find it much more difficult to make money off their work,

That doesn't look like it's the point to me.

""[the United States Congress shall have power] To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right , to their respective Writings and Discoveries." "

As I read that, copyright is there to 'promote progress', not to maximize gains.

No doubt there is a million linear feet of case law that got us where we are.

Honestly, I rather like this whole question of copilot. I solidly appreciate the brilliance of github as a honeypot.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#214

Post by blihp »

yashap wrote: post #154
blihp wrote: post #33

Anyone publishing anything on the Internet should expect this type of use case. If it is removed from github and republished via another site, there is absolutely nothing preventing another service/company from doing the exact same thing (or 'worse'... i.e. imagine a learning system that can actually understand the code) when scraping the alternative location. It's not unusual for bots to be among the most frequent v…

A bigger concern for many is that if you USE copilot, you’ll unintentionally copy code with licences that your company really, REALLY does not want to copy. For example, here’s copilot copying some very famous GPL code: https://twitter.com/mitsuhiko/status/1410886329924194309?s=2... And basically every software company avoids GPL like the plague, due to its strong copyleft conditions.

Sure, but that's a different end of the issue than I was referring to. I was pointing out that just taking code off of github wouldn't avoid the use case. Any published code from any public source is likely to eventually be used this way by someone.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#215

Post by kingofclams »

bsd44 wrote: post #57
lloydatkinson wrote: post #39

Earlier quoted context omitted.

Oh man I can’t imagine the consequences for certain languages and frameworks if it uses SO answers though. Imagine if it trained in all the dumb and ancient answers like “how do I get the length of a string in javascript” and took the first accepted answer of “use jquery”

This raises an issue of trolling. What prevents developers to generate "inappropriate" code to feed it to this algorithm the same way they did with the Microsoft Chat bot for example? That will surely reflect on the quality of code generated by this AI system and therefore the stability and security of applications built.

I’m sure this will happen, and there will definitely be instances of the bot giving users bad code, but it would be incredibly difficult to make it solely give out bad code.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#216

Post by IncRnd »

IfOnlyYouKnew wrote: post #179
IncRnd wrote: post #74

Earlier quoted context omitted.

> Nobody is really being hurt when a new tool makes it easier to copy little bits of code from the internet. That's the first time I've heard copilot get described as copying little bits of code from the Internet. Copilot aggregates all github source code, removes licences from the code, and regurgitates the code without licenses. Furthermore, both github and the programmers using copilot know this. Look at any one o…

It's not "literally" stealing, because it doesn't deprive anyone of the use the source code. Those two points were somehow extremely obvious to everyone here as long as it was music and movies we were talking about. And Github themselves have stated that only 0.1% of the Copilot output contains chunks taken verbatim from the learning set. Of those, the vast majority are likely to be boilerplate so generic it's silly…

> It's not "literally" stealing, because it doesn't deprive anyone of the use the source code.

That's simply not true. You might be confusing idealism about software freedom with how both law and society define theft.

Edit: In this comment I refer to the US.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#217

Post by saurik »

sillysaurusx wrote: post #134
hojjat12000 wrote: post #50

Earlier quoted context omitted.

Researchers in our lab created a huge dataset of facial expressions from images on the web, annotated it and published the URLs to the images and the annotations for research but made sure to search only for images with proper licenses. I don't think that you are allowed to just go download any old image and train on it. I understand the many many people do it, but it's not legal (as far as I know, please correct me…

> I don't think that you are allowed to just go download any old image and train on it. My understanding as a two-year student of ML is that you are allowed in the US to go download any old image, train on it, and then release the model as long as the outputs are "sufficiently transformative." That last phrase is the key part, and has never been tested in court. It's entirely possible that either I'm mistaken here, o…

To be clear: "transformative" not meaning merely "altered" but really meaning "repurposed"; if the new work is something people could feasibly use instead of the old work (harming the author's original market), it isn't "transformative".

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#218

Post by jozvolskyef »

saurik wrote: post #139
moyix wrote: post #124

Earlier quoted context omitted.

Have you used Copilot? I have not, but I have trained a GPT2 model on open source projects ( https://doesnotexist.codes/ ). It does not just pattern match and replicate. It can be cajoled into reproducing some memorized snippets, but this is not the norm; in my experience the vast majority of what it generates is novel. The exceptions are extremely popular snippets that are repeated many many times in the training da…

So, to verify, your claim is that GPT-3, when trained on a corpus of human text, isn't merely managing to string together a bunch of high-probability sequences of symbol constructs--which is how every article I have ever read on how it functions describes the technology--but is instead managing to build a model of the human world and the mechanism of narration required to describe it, with which it uses to write new…

Imagine for a second a model of the human brain that consists of three parts. 1) a vector of trillion inputs, 2) a black box, and 3) a vector of trillion outputs. At this level of abstraction, the human brain "pattern matches and replicates" just the same, except it is better at it.

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#219

Post by lucideer »

lacker wrote: post #34

I thought this was a pretty good thread (by an ex-Wikipedia lawyer) on Twitter about the IP meaning of Copilot. https://twitter.com/luis_in_brief/status/1410242882523459585... And this is a longer article about how IP and AI interact: https://ilr.law.uiowa.edu/print/volume-101-issue-2/copyright... I am not a lawyer, but I am capable of summarizing the thoughts of lawyers, so my take is that in general, fair use allow…

There's a lot of sibling commenters disagreeing with this take but I think they miss that ultimately this comes down to how legal experts interpret tech, rather than what tech experts think law should apply.

This is, imo, unfortunate, as often the legal interpretation is based on a gross misunderstanding of how the tech works, but this is the way.

I don't think copilot should be legal according to my own interpretation but in this (rare) case I feel the "IANAL" tag applies not because I lack (legal) knowledge, but rather because I have (tech) knowledge that is likely absent from actual decision making on legal outcomes (therefore leading to different legal outcomes than how I would see things working).

Re: I do not agree with Github's use of copyrighted code as training for Copilot

#220

Post by lolinder »

cjohansson wrote: post #190

It's strange how this thing was not an opt-in feature at GitHub. I also feel like this thing is a violation of my integrity and I will consider stop using GitHub as well

From their perspective, they weren't doing anything that abused their privileged position. Tabnine trained their model on open source code, much of which was probably hosted on GitHub. Why should GitHub have to ask permission if tabnine didn't?

Whether training an ML model on code is fair use is still an open question, but I don't think GitHub is a greater villain here than anyone else doing the same thing (at least until they start using private repos).

Post reply on HN