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

GitHub Copi­lot inves­ti­ga­tion

githubcopilotinvestigation.com

Post reply on HN
Print view View original

291–300 of 1001 posts

Re: GitHub Copi­lot inves­ti­ga­tion

#291

Post by mark_l_watson »

First, the author’s book Beautiful Racket is very cool, recommended.

I largely disagree with this article, at least for MIT, BSD, etc. training code examples. The small autocompletions, even if they are several lines long, sort of seems like fair use to me.

I do think that CoPilot should have an option to use a smaller model just trained in code that has very liberal use licenses, because I think the use of GPL, etc. licensed code is problematic - at least for me.

For what it is worth, I have a lot of Apache 2 licensed repos on GitHub (largely examples from my books) and I am pleased if my code contributed a small bit to the CoPilot training data. I also publish my recent books under Creative Commons, allow reuse, even commercially licenses: basically anything I do that might help someone, I am all in for sharing.

Re: GitHub Copi­lot inves­ti­ga­tion

#292

Post by larsiusprime »

BoppreH wrote: post #241

Most of these points can also be raised against DALL-E 2, but software has one extra thorn: patents. It's a common advice to not read software patents[1] because the infringement penalties are lower if you did so unwittingly, that is, by reinventing the patented technique yourself. I wonder if using Copilot doesn't push the penalties back again to wilful infringement. Or worse, patent trolls poisoning the training da…

Has Dall-E 2 yet reproduced 1:1 anything from its training set?

Re: GitHub Copi­lot inves­ti­ga­tion

#293

Post by 627467 »

Hey, I despise bait and switch from large corps. But I also find it unsustainable this idea that societies and legal resources are wasted fighting for IPs.

The code is out there. Millions of people are being trained and writing code based of the learnings of open data.

Designers have "mood boards". Developers have open source. Right now I don't have sympathy for MS, but in a few years any you developer could just do what MS is doing with Copilot in their bedroom. Why would you care about the kid in their bed room training an AI with free (as in public) information?

Re: GitHub Copi­lot inves­ti­ga­tion

#294

Post by ROTMetro »

nomagicbullet wrote: post #260
giobox wrote: post #148

Earlier quoted context omitted.

Its not popular to say it, but I agree to some extent here too. We may need a wholesale reimagining of copyright/patents in many places to accept the new reality of both of building the tools (data to train) and in accepting the occasional bad output (copyright/patented function appears in output). I think watching the law evolve with the tech is going to have a lot of ups and downs.

I think you hit the nail on the head. Our laws and rules were created for a cultural context that is quickly becoming outdated. I feel there are many valid criticisms of AI today, but demonizing the technology doesn't allow for fruitful discussions. We need to evolve our thinking and we need to be open minded to do so first.

What makes the rules outdated? The fact that you want to get away with what they were designed to prevent?

Re: GitHub Copi­lot inves­ti­ga­tion

#295

Post by thegrimmest »

treis wrote: post #190
beachy wrote: post #103

Earlier quoted context omitted.

Github does plenty of stuff that you can't do. e.g. the contributions graph as just one example that comes to mind. That's not unfair, and not the basis for a lawsuit, it's just business. > Copilot is not only making money off of open-source, they are making money off of open-source in a way others can't. Of course! That's why MS paid squillions to buy Github.

You're not supposed to be able to use dominance in one market (git hosting) to gain dominance in another (AI powered code suggestions).

But who gets to draw the lines? Why can't both of these be in a single market "tools for programmers"?

Re: GitHub Copi­lot inves­ti­ga­tion

#296

Post by batmanturkey »

deworms wrote: post #144
Sirened wrote: post #129

Earlier quoted context omitted.

You aren't allowed to just read code and regurgitate it in order to claim it as your own. That is, just because you memorized this great new novel you read, it doesn't mean you can go and sit down and hammer it out and sell new copies. People go to great lengths to do this sort of things (see: clean room reverse engineering [1]) in order to try and wash themselves of liability. [1] https://en.wikipedia.org/wiki/Clean…

If you think most people pay any attention to licenses or respect them you better think again. Snippets get copied verbatim with no regard to their source all the time. Licenses have no power and are routinely ignored.

Mmm maybe rephrase that as “depending upon which entity’s copyright was violated”

Surely I don’t need to recite the last 50 years of tech legal precedent and case history for you to see that such a blanket generalization cannot be left unaddressed.

Litigants litigate

Re: GitHub Copi­lot inves­ti­ga­tion

#298

Post by zo1 »

armchairhacker wrote: post #37

One issue I see with Copilot is that they get free access to all open-source data on GitHub, but using GitHub APIs to download the data yourself isn't possible (rate limiting). This is an unfair advantage. Copilot is not only making money off of open-source, they are making money off of open-source in a way others can't. I would love to see a lawsuit which requires GitHub to provide their full Copilot dataset.

I think the very least they could do is dump the list of URLs from which they took the source files. The community will figure out the rest.

Re: GitHub Copi­lot inves­ti­ga­tion

#299

Post by wilg »

andrewxdiamond wrote: post #282
spywaregorilla wrote: post #268

Earlier quoted context omitted.

Suppose you wanted to do what some code does, then you see this LGPL code. What can you do? Adjust variable names and play with line spacing and comments until it feels different?

Write it yourself

So, then, there's no way to know whether someone wrote it themself or "copied" it verbatim.

Re: GitHub Copi­lot inves­ti­ga­tion

#300

Post by TheMiddleMan »

There's a big difference between learning and memorizing.

If the AI is "learning" how it works by studying public code then using its knowledge to create, that's okay.

But if it's just memorizing code and reciting it back, not okay. Just like if a human were doing this.

Of course we don't currently have ways to know the difference [that I know of] since AI is a black box.

Interestingly, current AI is not capable of truly understanding how code works and how it will execute, so it has to learn in it's own way. I suspect it can learn what valid syntax is, but I doubt it is aware of how the code will execute.

It's possible this is just a case of Overfitting. https://en.wikipedia.org/wiki/Overfitting

Post reply on HN