Hacker News Reader

This is a really good point that I hadn't considered before. It's facebook all over again — selling your own content back to you. Repo owners should be at least compensated when their code gets used. That would be an incredible market.

> That would be an incredible market

I for one welcome our new CEO (Copilot Engine Optimization) overlords.

Jokes aside that will likely cause GitHub to be filled with lots of low quality repos (even AI generated, oh the irony!), to trick Copilot into using their code.

> Jokes aside that will likely cause GitHub to be filled with lots of low quality repos

Hasn't this already been the case since GitHub became a CV boost.

I don't think that would be possible. One of the big limitations of neural networks is that they don't cite their sources.

Calculate a rough semantic similarity score across all your snippets, and pay out a fractional reward to all originating codebases.

I think the bigger problem is that it will almost certainly lead to a proliferation of giant snippet spam repositories.

It is like the search engines vs. SEO arms race. The hope could be that such proliferation can be managed by disincentivizing such abuses. The reality might be vastly different with codes that have more regularity and better chance for AI's emulating humans than the natural language texts.

Then they need to negotiate a non-attributed contract with you before using your code to train (not sure abiut testing though).

It should be automatic based on license. GPL code definitely shouldn't be included but MIT could be. They already have this information in most repositories and if its missing they have no right to use it at all. We don't need extra options the licenses already restrict the use and derivative work.

Not without the text of the license. I, as a developer, cannot just poach open source code under MIT without including the copyright and terms from the original project. From the license:

"The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software."

They might argue that a snippet isn't a "substantial portion" of "the Software", and they're only charging for the service not the content - regardless, I don't like it, this is exactly what certain licenses attempt to prevent.

I would argue that substantial shouldn't be measured in lines of code, it should be measured in importance. Something like the fast inverse square root is substantial even though it's short.

The fast inverse square makes for a poor example when it comes to notions of intellectual property and copyright because there is prior art. The Wikipedia page has a history.

And imagine if Microsoft had been able to copyright the fast inverse square function before Carmack sat down to write Quake!

Prior art matters for patents, not copyrights. Carmack's code is still protected by copyright even if he didn't invent the algorithm.

And copyright doesn't prevent someone else from implementing the same algorithm, only from copying the code. If Microsoft had been able to copyright the fast inverse square root function, Carmack could still have written his own version and even copyrighted that version himself.

Prior art has been used for copyright in plenty of court cases, and increasingly so!

Here’s an example: https://scholar.google.com/scholar_case?case=728470765881077...

It seems particularly apt to consider prior art for use in software IP if only for the similarities with the patentable invention of mechanical parts.

What? You are free to reverse engineer any non patented object and reproduce an almost identical object. Prior art for copyright is meaningless unless it is about authorship. You can still do a clean room implementation and ignore prior art.

But the current interpretation is that you cannot claim authorship over things that were already in the public domain. Feel free to read the case notes and follow some of the links in there for more info. Am I not interpreting the court rulings correctly?

——

Considering de novo the evidence before the district court, we hold that the district court did not err in granting summary judgment. Johannsongs failed to offer admissible evidence to rebut Ferrara's analysis, so there is no genuine dispute of material fact as to his conclusions that Söknuður and You Raise Me Up are not substantially similar and most of their similarities are attributable to prior art. Based on these conclusions, Johannsongs has failed to satisfy the extrinsic test and Defendants are entitled to judgment as a matter of law.

but they used all your code to train. that's pretty substantial..

I too have reservations about Copilot, but does the MIT license define a "substantial portion"? I doubt a snippet would fall under either "copies" or "substantial portions"

I doubt many licenses define that kind of terminology. That's left to precedents established by actual cases. My point was just that you're not free to use code from an MIT-licensed project without following the terms of the license. The other details get worked out when legal actions are taken.

Well, I hope your viewpoint doesn't win the day, because making code as freely shareable and remixable as possible is a huge boon for humanity.

Code being freely shareable and remixable is great. Selling that open source code for profit is not.

Is your take that Microsoft should offer this for free? Or if they are not willing to do it for free, Microsoft should cancel this service and we should wait for Apache or someone else to offer the service?

Or something else ?

Microsoft should make this service free for open source (not just thought leaders), and compensate people otherwise. I should have a 0.01% equity in Open AI if they're using my stuff like this.

Or they should do opt in.

Half serious/flippant, we need MS to create a cryptocurrency so that developers can be credited with micropayments each time their code gets “quoted” in the IDE.

<ducks>

What is wrong with someone making a little dough. It is just numbers in database.

Yeah, but those numbers translate to food on the table for my kids, a roof over their heads, better education, etc. Come on, this is a tired response. Nothing is wrong with people making money. There is a lot wrong with people making money off of the hard work of others without any consideration or remuneration.

It's just as shareable on Gitlab, no? And the issue isn't that code is not shareable - it's that a huge corporation is profiting from this code without consent from the developer.

> a huge corporation is profiting from this code without consent from the developer

Also without attribution. The more permissive licenses allow corporations to profit from shared code, but most of them still require attribution.

And it's really not much to ask: when someone gives you free code, give them credit for their work.

Well, I hope your viewpoint doesn't win the day, because breaching GPL left and right to make some developers life easier opens huge cans of worms.

I feel like you're missing the forest for the trees here - making code freely shareable and remixable is exactly the purpose of GPL and other free-software licenses, but you can bet that the proprietary codebases Copilot will be used in will go out of their way to prevent any such uses of _their_ particular code snippets.

IMO, the only way to use Copilot's output in an ethically sound way is to only use the output it produces in AGPL licensed projects (assuming that Copilot has not been trained on any non-free software codebases which in itself is a strong assumption).

> IMO, the only way to use Copilot's output in an ethically sound way is to only use the output it produces in AGPL licensed projects

Even then, that is missing attribution which should really be the default for all code reuse and derivation even when you legally are allowed to omit it.

Based on this comment, you may not understand what the GPL's purpose actually is, because it is NOT simply for the promoting sharing and remixing. The GPL is for ensuring that code, and its derivatives, are all able to be shared and remixed in perpetuity. the biggest (imo) difference between GPL and MIT/BSD licenses is that you CAN NOT use GPL'd code in a non-GPL* codebase. (*or GPL-compatible license)

The Twitter thread’s title seems unnecessarily incendiary and clickbaity.

I don’t buy that producing/synthesizing code snippets based off public repos is a problem.

There’s nothing proprietary or original about eg. the syntax of a for-loop, or the boilerplate of setting up some JS framework MVC.

Besides, it’s basically just a (semantic and contextual) search engine inlined within the IDE. Copyright infringement hasn’t taken place until the user activated the autocompletion and actually placed the code within their own and released their code containing the infringing code.

> There’s nothing proprietary or original about eg. the syntax of a for-loop, or the boilerplate of setting up some JS framework MVC.

Of course there is something proprietary or original about that. Why else would they need such an enormous AI to suggest it. Auto completing simple boilerplate was already solved in a much simpler way.

> Copyright infringement hasn’t taken place until the user activated the autocompletion and actually placed the code within their own and released their code containing the infringing code.

Copyright infringement takes place as soon as some company publishes/sells material without explicit license or permission. So not the moment the users hits accept, but the moment just before that: when the tool shows it to the user.

Applying your logic, is any search engine infringing on copyright because it contains a snippet of the source page?

After all, if showing a search result in the IDE is “publishing” (let alone “selling” (?)) why hasn’t Google been sued out of existence for showing search results (oops, “publishing” copies of original work, billions of times over), as well as selling related advertising?

https://en.wikipedia.org/wiki/Fair_use

> Examples of fair use in United States copyright law include commentary, search engines, criticism, parody, news reporting, research, and scholarship.

is any search engine infringing on copyright because it contains a snippet of the source page?

In some jurisdictions, it is. And in other jurisdictions, it is only allowed as long as it shows a link to the source page, which Copilot also doesn't do.

I stopped publishing open source after all this started coming out because I was so uncomfortable with it.

If your code is using a license that allows it, how could you possibly opt-out aside from using a different license?

Microsoft could provide an opt-out for projects or even contributors, regardless of licence.

> If your code is using a license that allows it, how could you possibly opt-out aside from using a different license?

A repo setting that instructs Github not to use your code for Copilot, it could be a similar option as turning Discussions on / off.

If they really want to win developers over they would even have Copilot scanning disabled by default but that'll never happen.

Sounds like you want a new license that just prohibits use by one company for one purpose.

There are other AI-based code completion systems than Copilot, at least Tabnine[1] and Kite[2] come to mind, I'm sure there are more.

[1] https://www.tabnine.com/

[2] https://www.kite.com/

As of today there is a new one...

"Now in Preview – Amazon CodeWhisperer"

https://aws.amazon.com/blogs/aws/now-in-preview-amazon-codew...

is there something wrong with that?

Not at all, you can put any license on your code that you want.

Even if Github did provide that setting, as a courtesy, someone could clone / fork the code to another repo (if you use any licence that allows it) and not enable that setting.

Sure that's possible, but there's a huuuge difference between Possible and Default Behavior.

In a case like this, GitHub itself could set up a bot account that forks all projects as soon as you make the switch. The company in fact would be incentivized to do so.

Does GitHub verify that the code that is in my repository is actually in accordance to the license that I’ve added? I could just upload any proprietary code with an incorrect license, and GitHub would just use that to feed their AI. Like any other dependency that you incorporate into your application, GitHub should verify/audit whether the license allows them to do so.

Don’t most licenses require at least attribution? I don’t believe GitHub is restricting themselves to only licenses that don’t. In fact the only software licenses I can think of that don’t require attribution are 0BSD, WTFPL, CC0, MIT-0 and Unlicense, and they all aren’t super popular. Also in some countries creators have inalienable moral rights which can be enforced regardless of the license. For example in Germany it is impossible to relinquish certain rights you have as the creator of a work, including the right to attribution.

This is an important and overlooked point. Even common permissive licenses (ISC / MIT / Apache-2.0) require attribution

Just as a mind experiment: couldn't CoPilot just publish a list of every github user and attribute the work to all of them?

CoPilot is a black box at the moment. Microsoft claims they used the public corpus on GitHub. There are plenty of GPL, AGPL, and "source available" projects in the public corpus. So what exactly is the licensing?

The argument may make sense if they limited themselves to public-domain (CC0) works, but that is not what happened here. If CoPilot attributed something to an AGPL project, does it mean the "virality" applies to all projects that use code from CoPilot?

There's also a good amount of commercial and leaked source code on GitHub, including MS's own leaked Windows XP source. I haven't played around with Copilot yet, but if I ever do I plan on copy/pasting some win32 API definitions to see if I can get it to spit out any of the leaked source.

> if I ever do I plan on copy/pasting some win32 API definitions to see if I can get it to spit out any of the leaked source.

If that works, then I can't wait for that to be a boon for Wine and ReactOS: "Microsoft itself provided this code and allowed us to use it, so therefore it's totally legal. Neener neener."

Some trivia: CC0 is a public domain declaration.. at least in the US. There is no process by which an author can make their works public domain, CC0 is just a (weak) promise that the copyright holder will treat the work as if it were public domain.

This feels like a tool that can easily be destroyed by a lawsuit, I can't imagine a TOS can force you to give away your copy rights (especially if they allow and encourage you to post your own copyright).

If it can't then Wikipedia is doomed; its entire licensing status rests on the notion that editors grant such a license as part of their clickwrap ToS.

I'm not sure using a different license actually opts you out. By merely hosting your code on GitHub you grant them the right to analyze your code on their servers[1]

They may be morally in the wrong, but I'm unsure they are legally in the wrong here. To boot, denying them the right to create this tool in your license is technically a violation of OSS principles and problematic

[1]: https://docs.github.com/en/site-policy/github-terms/github-t...

> This license does not grant GitHub the right to sell Your Content. It also does not grant GitHub the right to otherwise distribute or use Your Content outside of our provision of the Service, except that as part of the right to archive Your Content, GitHub may permit our partners to store and archive Your Content in public repositories in connection with the GitHub Arctic Code Vault and GitHub Archive Program.

Wouldn't this be a violation?

It would be kind of cool if Github could show some stat that code you wrote has been used 50,000 times for 12,000 people.

Being a top CoPilot contributor should at least have value to signal on your resume.

> This might be overreacting but is there a way to opt-out of Copilot using your code in open source repos?

I don't think there is a way to opt out if it is a public repo regardless of license, and Microsoft's copyright theory suggests that they wouldn't feel obligate to enxclude any code they got their hands on except under a specific NDA preventing such use; the use of public GitHub repos isn't based on legal constraints but practical convenience.

They’re not selling you code. They’re selling you an engine that helps you find the right free code at the right time.

If you read free code yourself it’s fine, but if a machine does it for you it’s not? We overvalue humans.

> If you read free code yourself it’s fine, but if a machine does it for you it’s not? We overvalue humans.

No, it's not fine. Apparently, you missed SCO & Oracle vs. Google cases. Both of these cases argued that somebody looked to the code, and copied it. In SCO case it was not true, but the argument stretched the timeline rather successfully. In Oracle vs. Google, copying function signatures opened a big can of worms.

So, just by copying the function signature without filling it the very same code with the original, even for interoperability, you're getting into a huge gray area in a legal sense.

Similarly, no sane Wine developer will read leaked Microsoft source code, yet alone copy it. Again, no sane emulator developer will read leaked Nintendo code.

Reading the code "colors" your creativity, and if you're tried at court and enough similarity is found in your code with the leaked code, it's game over.

So, reading code and copying is not guaranteed to be legal, depending on its license. When this is done by a robot, it's still illegal (you're breaching licenses during the code generation process), and immoral and unethical on top of it.

So, we don't overvalue humans, but overvalue AI, which is just informed search, BTW.

I suppose you learned to code without reading any code?

No, I didn't and don't read other people's code to understand how something works. I use books and official language/library documentation for that.

On the other hand, this is irrelevant to the issue at hand.

GitHub copilot is not a tool for education. It's tool for auto-completing code, which can be put to production, where licenses and other stuff come into play.

The issue is not code sharing per se. It's more of a legal problem, and an important one at that. In the software copyright sense, even reading code you can't import to a project (let it be leaked, not compatibly licensed or for any reason), puts you at risk of legal troubles. This is why we have methodologies like "clean room development".

In Copilot's case, you're possibly deriving a code from a source which contains many licenses, and some of them are not compatible with that you're doing. As a result, you're in direct breach of the license which is not compatible with your code.

On a more higher level, you're also breaching the ethics code and morality by using a code or its derivation with an incompatible license to your code, and disregarding other peoples desires codified as a case-tested and valid license.

As a result, if you think that using a derivative of a GPL licensed code in your closed source application is OK on every front, then the vice versa is true. I can disassemble and reverse every part of your code and re-implement it as GPL bug for bug and open it.

Because if you can breach my license, and expect no consequences, I can breach your license without consequences, as well. It's a two way street.