Disallow or discourage use of "AI" tools (Christoph Cullmann)

[Christoph Cullmann]

Hi,

> 

> 

> On Sonntag, 18. Mai 2025 16:52:00 Mitteleuropäische Sommerzeit Christoph
> Cullmann wrote:
> 

> > On Sunday, May 18th, 2025 at 09:12, Albert Vaca Cintora
> 

> albertvaka at gmail.com wrote:
> 

> > > On Sun, 18 May 2025, 08:59 Justin Zobel, justin at 1707.io wrote:
> > > 

> > > > If the contributor cannot tell you the license(s) of the code that was
> > > > used to generate the code, then it's literally gambling that this code
> > > > wasn't taken from another project by Gemini and used without their
> > > > permission or used in a way that violates the license and opens up the
> > > > KDE e.V. to litigation.>
> > > 

> > > I'm no lawyer but I would expect that training AI will fall under fair use
> > > of copyrighted code. If that's not the case already, it will probably be
> > > soon. The benefits of AI to society are too large to autoimpose such a
> > > roadblock.
> > 

> > if that would happen, then there is just no copyright protection anymore and
> > all is fair game, I highly doubt that, but yes, that is what companies that
> > want to get rich with deep learning want to have.
> 

> 

> But isn't that how knowlegde transfer works since ages especially in science?
> 

> I learned programming mostly be reading other people's code. Have I been
> violating copyright or licenses for decades because I applied the patterns I
> saw in other Free Software code to my code? And what about stuff I look(ed) up
> on stackoverflow? I don't think I have ever seen code on stackoverflow that had
> a proper license. At least in Germany this means that this code is under a
> very strict license and cannot be used without the authors explicit consent.
> (Something like Fair Use doesn't exist in Germany.) On the other hand, I don't
> think that I have ever copied code literally from stackoverflow. I have also
> never copied code literally that I developed at my former workplace, but of
> course I have applied some of the concepts I learned from writing proprietary
> code to my Free Software code.
> 

> Yes, there is the theoretical threat that an AI learned code that's under a
> less liberal license like the GPL or even under one of the "new" not-OSI-
> approved licenses used by certain companies to prevent Amazon from selling
> services based on their code (or even proprietary code; for all we know, Co-
> Pilot was trained with the entire source code written by Microsoft), but
> that's only a problem if the AI cites this code literally so that we could be
> sued for plagiarizing. How realistic is this threat?

I think the big issue is: you don't know.

That is all just deep learning, the model doesn't give you a clue if it
just derived something based on combinations of some 'learned' variants
or just outputs more or less the exact article/snippet/... it was trained on
with some minor renamings.

On the other side I would assume, if you do create a code snippet,
it will not be by accident a full copy of something you have seen somewhen,
or you will at least tell that :)

But back to my

https://invent.kde.org/frameworks/syntax-highlighting/-/merge_requests/698

as example: I can live with ignoring the model and just say: the user put in
the referenced:

rust.xml - MIT
bash.xml - LGPL
powershell.xml - MIT
https://docs.kde.org/stable5/en/kate/katepart/highlight.html
A code a previously worked as a basis to test against: https://github.com/imsys/fzf/blob/master/shell/completion-external.nu

and then accept it as LGPL.

Is that ok? For me it would.

Greetings
Christoph

> 

> Regards,
> Ingo
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 343 bytes
Desc: OpenPGP digital signature
URL: <http://mail.kde.org/pipermail/kde-devel/attachments/20250518/04ffb21b/attachment-0001.sig>