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

All public GitHub code was used in training Copilot

twitter.com

Post reply on HN
Print view View original

641–650 of 734 posts

Re: All public GitHub code was used in training Copilot

#641

Post by singhkays »

mceachen wrote: post #604
mikewarot wrote: post #575

Earlier quoted context omitted.

I have about 300,000 photos that haven't been scanned by AI (unless someone at Backblaze did it without permission). I'm sure there are lots of other photographers out there who miss Picassa, which Google killed off to push everyone's data to their service. (It did really well in matching faces, even across age, but the last version has a bug when there are multiple faces in a picture, sometimes it swaps the labels)…

Absolutely possible. What I'm building into PhotoStructure is typically called "transfer learning." https://en.wikipedia.org/wiki/Transfer_learning PhotoStructure is entirely self-hosted, including model training and application: the public domain base models (trained on huge datasets) are fetched and cached locally. By design, none of your data (or even metadata) leaves your server. (I expect to ship this in an upco…

Nice! Will try this out. Are you planning on taking advantage of in-built neural engines like that in Apple M1 for speeding up object/facial recognition?

Re: All public GitHub code was used in training Copilot

#642

Post by fergie »

habitue wrote: post #353

I am really confused by HN's response to copilot. It seems like before the twitter thread on it went viral, the only people who cared about programmers copying (verbatim!) short snippets of code like this would be lawyers and executives. Suddenly everyone is coming out of the woodworks as copyright maximalists? I know HN loves a good "well actually" and Microsoft is always suspect, but let's leave the idea of code la…

If very powerful companies are appropriating and reproducing code in contravention of copyright then that is something that should be called out.

Re: All public GitHub code was used in training Copilot

#643

Post by colordrops »

mikewarot wrote: post #575
fleddr wrote: post #551

To me, the particular use case and whether it is fair use or not, is of minor interest. A far more pressing matter is at hand: AI centralization and monopolization. Take Google as an example, running Google Photos for free for several years. And now that this has sucked in a trillion photos, the AI job is done, and they likely have the best image recognition AI in existence. Which is of course still peanuts compared…

I have about 300,000 photos that haven't been scanned by AI (unless someone at Backblaze did it without permission). I'm sure there are lots of other photographers out there who miss Picassa, which Google killed off to push everyone's data to their service. (It did really well in matching faces, even across age, but the last version has a bug when there are multiple faces in a picture, sometimes it swaps the labels)…

You don't encrypt your data before uploading to backblaze?

Re: All public GitHub code was used in training Copilot

#644

Post by colordrops »

pyuser583 wrote: post #633
fleddr wrote: post #551

To me, the particular use case and whether it is fair use or not, is of minor interest. A far more pressing matter is at hand: AI centralization and monopolization. Take Google as an example, running Google Photos for free for several years. And now that this has sucked in a trillion photos, the AI job is done, and they likely have the best image recognition AI in existence. Which is of course still peanuts compared…

Google doesn’t have the best (publicly) available reverse image search AI. That would be Yandex. Google is actually pretty crappy at reverse image searches. https://www.bellingcat.com/resources/how-tos/2019/12/26/guid...

It's pretty clearly intentionally hobbled for various reasons (e.g. privacy, obscenity, etc). It used to work a lot better.

Re: All public GitHub code was used in training Copilot

#645

Post by GuB-42 »

My guess is that is is fair use but...

The model is said to spit out code verbatim 0.1% of the time, a low number, but if copilot is used a lot, it means you are going to find a lot of copied code in people's projects, and these project owners may be breaching copyright. I don't think "but, copilot..." will be an excuse.

Here is a (probably unrealistic) scenario illustrating it. I am playing a copyright troll here:

- Release plenty of generic code and put it on GitHub under a restrictive license

- Have the copilot bot scan it

- wait some time

- scan public codebases and do an exact match for my code

- sue project owner that contain my code

I see the use of copilot more of a minefield for me than as a liability for Microsoft.

Re: All public GitHub code was used in training Copilot

#646

Post by IfOnlyYouKnew »

colordrops wrote: post #644
pyuser583 wrote: post #633

Earlier quoted context omitted.

Google doesn’t have the best (publicly) available reverse image search AI. That would be Yandex. Google is actually pretty crappy at reverse image searches. https://www.bellingcat.com/resources/how-tos/2019/12/26/guid...

It's pretty clearly intentionally hobbled for various reasons (e.g. privacy, obscenity, etc). It used to work a lot better.

Which would explain why Yandex, specifically, is the best-in-category: being based in a country where your government enjoys trolling developed world's idea of decency and responsibility can have its advantages.

(Until, of course, they force you out and give company to some crony oligarch. But that idea is also not unknown to Yandex, I believe?)

Re: All public GitHub code was used in training Copilot

#647

Post by bayindirh »

mikewarot wrote: post #575
fleddr wrote: post #551

To me, the particular use case and whether it is fair use or not, is of minor interest. A far more pressing matter is at hand: AI centralization and monopolization. Take Google as an example, running Google Photos for free for several years. And now that this has sucked in a trillion photos, the AI job is done, and they likely have the best image recognition AI in existence. Which is of course still peanuts compared…

I have about 300,000 photos that haven't been scanned by AI (unless someone at Backblaze did it without permission). I'm sure there are lots of other photographers out there who miss Picassa, which Google killed off to push everyone's data to their service. (It did really well in matching faces, even across age, but the last version has a bug when there are multiple faces in a picture, sometimes it swaps the labels)…

> If there were offline image recognition we could train on our own data privately...

Apple does all face recognition and image processing stuff on the edge. On your iPhone or Mac.

I wondered why my phone got frighteningly hot while charging sometimes. Then I saw the note after adding some faces manually for it to recognize, which was in the line of "Your phone will update faces when the phone is charging". My all photos are backed up to iCloud, btw.

Re: All public GitHub code was used in training Copilot

#648

Post by LouisSayers »

anigbrowl wrote: post #610
LouisSayers wrote: post #533

Earlier quoted context omitted.

> Perhaps people on HN start sensing that successors of Github Copilot will take their programming job. Rightly so. I feel like this comment misunderstands what a software developer is doing. Copilot isn't going to understand the underlying problem to be solved. It's not going to know about the specific domain and what makes sense and what doesn't. We're not going to see developers replaced in our lifetime. For that…

The thing is that understanding the domain and thinking out a fairly efficient or elegant solution is something a lot of industry specialist and scientists can do, and only part of programming. Another part is dealing with all the language syntax and specialist lego bits/glue code, and that's something domain specialists tend to be less good at and not enjoy spending time on; it's its own craft. Having a semi-intelli…

The funny thing with "good enough" solutions is that at some point it becomes unmanageable. I've basically spent a good part of my career cleaning up these solutions to make way for scalable, maintainable solutions that don't introduce security holes.

Photography is a good analogy - with everyone having fancy cameras you could think that a photographer is now not necessary. But yes there are still photographers about - they see things that the average person doesn't. The camera doesn't tell them what type of photos to take, what composition the photo should have or what poses a model should have.

Re: All public GitHub code was used in training Copilot

#649

Post by zohch »

remram wrote: post #559
kune wrote: post #550

Guys please read the Terms of Use of Github section D.4. We need the legal right to do things like host Your Content, publish it, and share it. You grant us and our legal successors the right to store, archive, parse, and display Your Content, and make incidental copies, as necessary to provide the Service, including improving the Service over time. This license includes the right to do things like copy it to our dat…

If I upload somebody else's GPL code to GitHub, I also can't grant to GitHub the (implicit) legal rights to use that code in Copilot, because they are not mine to give. I could previously mirror GPL code, because the GPL granted me the rights I need to grant GitHub as part of their ToS; but if they change their ToS, or if the meaning is changed by them adding vastly different features to their Service, this becomes a…

Can you explain what limitation in GPL would prevent someone from using it as training data? Also, if you are not allowed to upload GPL to GitHub, seems like the right answer is don't.

Re: All public GitHub code was used in training Copilot

#650

Post by BlueTemplar »

mikewarot wrote: post #615
Mehdi2277 wrote: post #598

Earlier quoted context omitted.

I work in the field. Transfer learning helps get you decent/good models, but the best models remain ones trained on large amounts of data. You may be able to get away with good performance and not great on your task. For some areas you really care a lot about long tail performance (like self driving) that you will need massive dataset. For other areas if your goal is to be the best relative to other large companies y…

If there are hundreds of people with 100,000 photos each, that collectively is a massive training database, with a lot more labels and diversity of subjects. By keeping the training data itself private, distributed and outsourced, you might be able to get otherwise unachievable levels of performance.

This isn't going to solve the data ownership issues though, since they contaminate the program trained on them (and its blackbox nature only makes it worse)... though I guess that specifically for copyright it's going to depend on the final usage of that tool ?
Post reply on HN