771–780 of 824 posts
Earlier quoted context omitted.
They know now.
copilot only generate copyrighted when it seen the code many many times and that called memorization in machine learning, machine learning researchers will always try to decrease the amount of memorization in their artificial neurons
Earlier quoted context omitted.
In some ways all these AIs are plagiarizing... I think creators should opt-in to ai models, as no current license was developed with this in mind.
Maybe its time for Creative Commons License to address this. I'm curious if No-Derivative would already prohibit this? Does the ND language need tweaking? Or do they need a whole new clause. Edit: I guess they do address it in their faq and I'd summarize it "Depends if copyright law applies and depends if it's considered derivative". https://creativecommons.org/faq/#artificial-intelligence-and...
Crackpot Theory: Copilot (and by association many ML tools) is a form of probabilistic encryption. Once encoded, it's virtually impossible to pull the code (plaintext) directly out of the raw ML model (the cyphertext), yet when the proper key is input ('//sparse matrix transpose'), you get the relevant segment of the original function (the plaintext) back. We've even seen this with stable diffusion image generation,…
Anybody looking at the source image and the generated result would say they are the same.
Consider each repo on github to be a movie. What copilot does is to search for sequences of frames from any movie which line up to create a new coherent movie. Individually, each frame is protected by the copyright of the movie it belongs to. But what happens if you take a million frames from a million different movies and just arrange them in a new way? That's the core question here. Is the new movie a new copyright…
I think it's a great time to explain why this won't hit AI art such as Stable Diffusion, even if GitHub loses this case. The crux of the lawsuit's argument is that the AI unlawfully outputs copyrighted material . This is evident in many tests with many people here and on Twitter even getting verbatim comments out of it. AI art, in the other hand, is not capable of outputting the images from its training set, as it's…
Want to say that again?
Can someone explain to me Microsoft’s decision here to use GPL code in the training set? It would seem like sticking to non-attribution / non-viral licenses would have kept them in the clear. Was that an insufficient size data set?
Did Microsoft use the source code of Windows (in whole or in part) as training input to Copilot?
I think if this is successful it will be very bad for the open world. Large platforms like github will just stick blanket agreements into the TOS which grant them permission (and require you indemnify them for any third party code you submit). By doing so they'll gain a monopoly on comprehensively trained AI, and the open world that doesn't have the lever of a TOS will not at all be able to compete with that. Copilot…
I hope this case will fail and establish a good precedent for all future AI litigations and may be even prevent new ones. Your code is open source - irregardless of license, one might read it as a text book and then remember or even copy snippets and re-use this somewhere else unrelated to the original application. If you don't like this, don't make your code open source. This was happening and is happening independe…
> irregardless of license Hard no. Please stop using open source code if this is how you think of it. Without licenses being respected, we don't get open source communities.
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…