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

Copilot sells code other people wrote

twitter.com

Post reply on HN
Print view View original

101–110 of 838 posts

Re: Copilot sells code other people wrote

#101

Post by afiori »

xupybd wrote: post #67
alpaca128 wrote: post #45

Earlier quoted context omitted.

The first can be automated without ML though. And once you use ML you cannot guarantee it won't copy-paste existing code. This whole thing would be fine if GitHub hadn't just used all public code on their platform, ignoring all involved licenses.

It changes the code for use. I'm not sure it can be considered a copy. It much like reading someone else's code and drawing ideas and patterns from that code.

Copyright sensitive environments are very careful not to do that.

Re: Copilot sells code other people wrote

#102

Post by alpaca128 »

rob74 wrote: post #56
alpaca128 wrote: post #45

Earlier quoted context omitted.

The first can be automated without ML though. And once you use ML you cannot guarantee it won't copy-paste existing code. This whole thing would be fine if GitHub hadn't just used all public code on their platform, ignoring all involved licenses.

The problem is, if they had used only code with a license that allows copying without attribution, there wouldn't have been a lot of code left...

Difficulty doing something legally doesn't justify breaking the law.

Re: Copilot sells code other people wrote

#103

Post by trention »

coldtea wrote: post #98
Agamus wrote: post #75

Earlier quoted context omitted.

This idea has been around for a while - why... "pretty soon"? And I'm sure I couldn't disagree with you more. Or are 'influence' and 'theft' the same now?

> Or are 'influence' and 'theft' the same now? They have been the same for most of history. People could openly copy titles, plots, parts, phrases, etc from prior work. Same for mechanical designs. The only thing preventing them was obscurity (e.g. the inventor trying to make it hidden) not any law or ethical idea that it's bad (there wasn't any). That's how things from math to gears to tunes got better (or changed o…

Except that AI will not lead to "golden ages of art production" because nobody gives a sh*t about art created by AIs. And nobody will.

Re: Copilot sells code other people wrote

#104

Post by alpaca128 »

xupybd wrote: post #67
alpaca128 wrote: post #45

Earlier quoted context omitted.

The first can be automated without ML though. And once you use ML you cannot guarantee it won't copy-paste existing code. This whole thing would be fine if GitHub hadn't just used all public code on their platform, ignoring all involved licenses.

It changes the code for use. I'm not sure it can be considered a copy. It much like reading someone else's code and drawing ideas and patterns from that code.

It has been shown often enough that Copilot can reproduce exact copies of snippets.

Re: Copilot sells code other people wrote

#105

Post by wolframhempel »

When my last company got acquired, part of the due diligence process was a scan of our codebase for snippets from stack overflow. Every snippet found that wasn't posted with a clear license by the author was challenged and we rewrote it.

Now, I'm not entirely sure how necessary this was from a legal perspective. But introducing an AI into the mix will bring up a lot of uncertainty when it comes to how much change is required for something to no longer be considered a copy/derivative.

Re: Copilot sells code other people wrote

#106

Post by danamit »

The code Copilot suggest from any given project most of the time is not enough to credit such project, when I look up code in some GitHub repo, and copy it fully or part of it, I do not credit that project.

I do not see Copilot as useful anyway.

Re: Copilot sells code other people wrote

#107

Post by happymellon »

FeepingCreature wrote: post #96

All I can think of is Steve Yegge [1]: "They have no right to do this. Open source does not mean the source is somehow 'open'." My code is on Github so that people can read it, reuse it and learn from it. "The freedom to study how the program works", as the FSF says. If some of the people reading it are machines, why would that matter? [1] http://steve-yegge.blogspot.com/2010/07/wikileaks-to-leak-50...

Because a lot of this code would be put into closed source software, which is against the licence and would prevent people from exercising the right to study how a program works.

Re: Copilot sells code other people wrote

#108

Post by afiori »

alpaca128 wrote: post #55
bartq wrote: post #36

Earlier quoted context omitted.

Other thing I'm worried about: how to retract facts from ML model? I guess it's impossible, you need to retrain from scratch with part X removed from training set. Or... people could invent layered ML models similar to docker - each layer would be marked what data it was trained with. Then at least you'd have some cache of trained model to re-use in next training session. Nasty stuff.

Or instead of inventing complicated layered ML models Github could just use each repo's license information to decide what's okay to use. Detecting licenses is already a feature on that site.

Many licenses requite attribution, which would be hard to track.

Re: Copilot sells code other people wrote

#109

Post by pornel »

Tough pill to swallow. Microsoft's actions don't seem fair, but fighting them with copyright could weaken fair use:

https://felixreda.eu/2021/07/github-copilot-is-not-infringin...

There's a good argument that demanding copyright protections on scraped datasets and short snippets is a double-edged sword. It could harm search engines, distribution of news, and non-commercial ML research too.

Re: Copilot sells code other people wrote

#110

Post by dmortin »

wolframhempel wrote: post #105

When my last company got acquired, part of the due diligence process was a scan of our codebase for snippets from stack overflow. Every snippet found that wasn't posted with a clear license by the author was challenged and we rewrote it. Now, I'm not entirely sure how necessary this was from a legal perspective. But introducing an AI into the mix will bring up a lot of uncertainty when it comes to how much change is…

Did the scan find the process if they changed the variable names, for example? Or is that considered a differing snippet then?
Post reply on HN