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

We've filed a lawsuit against GitHub Copilot

githubcopilotlitigation.com

Post reply on HN
Print view View original

71–80 of 824 posts

Re: We've filed a lawsuit against GitHub Copilot

#71

Post by ridiculous_fish »

ISL wrote: post #29

Can anyone with Copilot access give a short summary of its response to the prompts: function force=Gmmr2Array(mass1, mass2) and function [force, torque]=pointMatrixGravity(array1,array2) ? I'd love to know if some of my GPL v3 code [1, 2] has landed in the training set [1] https://github.com/4kbt/NewtonianEotWashToolkit/blob/master/... [2] https://github.com/4kbt/NewtonianEotWashToolkit/blob/master/...

For Gmmr2Array: https://gist.github.com/ridiculousfish/9a25f5f778d98ecd81099...

For pointMatrixGravity: https://gist.github.com/ridiculousfish/af05137a4090e92de3a97...

Re: We've filed a lawsuit against GitHub Copilot

#72

Post by avian »

dmitrygr wrote: post #7
thesuperbigfrog wrote: post #5

How original is the generated code? Can the generated code be traced back to the code used for training and the original copyrights and licenses for that code? If so, what attribution(s) and license(s) should apply to the generated code?

They demonstrate generated code being identical to some training code.

There were well known examples of copilot reproducing exact code snippets well before this lawsuit (e.g. the Quake's fast inverse square root function). Microsoft dealt with them by simply adding the offending function names to a blocklist.

In other words, if your open source project doesn't have such immediately recognizable code and didn't cause a shitstorm on Twitter, chances are copilot is still happily spewing out your exact code, sans the copyright and license info.

Re: We've filed a lawsuit against GitHub Copilot

#73

Post by m00x »

dmitrygr wrote: post #7
thesuperbigfrog wrote: post #5

How original is the generated code? Can the generated code be traced back to the code used for training and the original copyrights and licenses for that code? If so, what attribution(s) and license(s) should apply to the generated code?

They demonstrate generated code being identical to some training code.

Just like developers have never copy-pasted code from stack overflow or Github :):):)

Re: We've filed a lawsuit against GitHub Copilot

#74

Post by cothrowaway88 »

Made a throwaway since I guess this stance is controversial. I could not care less about how copilot was made and what kind of code it outputs. It's useful and was inevitable.

I'm 1000% on team open source and have had to refer to things like tldrlegal.com many times to make sure I get all my software licensing puzzle pieces right. Totally get the argument for why this litigation exists in the present.

Just saying in general my friends I hope you have an absolutely great day. Someone will be wrong on the internet tomorrow, no doubt about it. Worry about something productive instead.

This one has the feel of being nothing more than tilting at windmills in the long run.

Re: We've filed a lawsuit against GitHub Copilot

#75

Post by mkeeter »

IceWreck wrote: post #34

I am not against this lawsuit but I'm against the implications of this because it can lead to disastrous laws. A programmer can read available but not oss licensed code and learn from it. Thats fair use. If a machine does it, is it wrong ? What is the line between copying and machine learning ? Where does overfitting come in ? Today they're filing a lawsuit against copilot. Tomorrow it will be against stable diffusio…

Wine literally bans contributions from anyone that has seen Microsoft Windows source code:

https://wiki.winehq.org/Developer_FAQ#Who_can.27t_contribute...

Re: We've filed a lawsuit against GitHub Copilot

#77

Post by bugfix-66 »

m00x wrote: post #64
bugfix-66 wrote: post #28

Ask HN: I want to modify the BSD 2-Clause Open Source License to explicitly prohibit the use of the licensed software in training systems like Microsoft's Copilot (and use during inference). How should the third clause be worded? The No-AI 3-Clause Open Source Software License Copyright (C) All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided tha…

Get a lawyer since this is nonsense.

It's literally the standard BSD 2-Clause License, word for word, with an additional third clause:

  3. Use in source or binary forms for the construction or operation
     of predictive software generation systems is prohibited.
Hardly nonsense, but obviously you aren't equipped to judge. More about the BSD licenses:

https://en.m.wikipedia.org/wiki/BSD_licenses

Re: We've filed a lawsuit against GitHub Copilot

#78

Post by m00x »

ketralnis wrote: post #46
abouttyme wrote: post #17

I suspect this will be the first of many lawsuits over training data sets. Just because it is obscured by artificial neural networks doesn't mean it's an original work that is not subject to copyright restrictions.

Yeah yeah my code produces the complete works of Micky Mouse but it's it's okay because _algorithms_!

Copyright is different than patent law and license law.

Re: We've filed a lawsuit against GitHub Copilot

#79

Post by googlryas »

Swizec wrote: post #26
dmitrygr wrote: post #7

Earlier quoted context omitted.

They demonstrate generated code being identical to some training code.

How many ways are there to write many of the basic algorithms we all use though? Can I copyright "({ item }) => {item.label} "? Because I sure have seen that exact code written, from scratch, in many many places. I guess my question boils down to "What is the smallest copyrightable unit of code?" . Because I'm certain suing a novelist for copyright infringement on a character that says "Hi, how are you?" would be con…

No specific sources to provide, but a lot of analyses were written about this question regarding the Google v Oracle java API lawsuit.

Re: We've filed a lawsuit against GitHub Copilot

#80

Post by ridiculous_fish »

jmcphers wrote: post #53
ISL wrote: post #29

Can anyone with Copilot access give a short summary of its response to the prompts: function force=Gmmr2Array(mass1, mass2) and function [force, torque]=pointMatrixGravity(array1,array2) ? I'd love to know if some of my GPL v3 code [1, 2] has landed in the training set [1] https://github.com/4kbt/NewtonianEotWashToolkit/blob/master/... [2] https://github.com/4kbt/NewtonianEotWashToolkit/blob/master/...

First prompt: var G = 6.67e-11; var force = G * mass1 * mass2 / distance * distance; return force; Second prompt: % pointMatrixGravity - Calculates the gravitational force and torque on a % point mass due to a matrix of point masses. % % [force, torque]=pointMatrixGravity(array1,array2) % % Inputs: % array1 - 3xN array of point masses % array2 - 3xN array of point masses % Outputs: % force - 3x1 array of gravitationa…

FYI if you use the "Open GitHub Copilot" command in VSCode you will get up to 10 different outputs for the same prompt.

Intereting that my results were different than yours!

Post reply on HN