Re: Concerns/questions around Software Heritage Archive

[Olivier Dion <olivier.dion@polymtl.ca>]

On Mon, 18 Mar 2024, Kaelyn <kaelyn.alexi@protonmail.com> wrote:
> On Monday, March 18th, 2024 at 2:28 AM, Simon Tournier <zimon.toutoune@gmail.com> wrote:

[...]

>> That’s the double sword of “free software”. :-)
>
> Hi,
>
> I want to stress that I am not a lawyer, but my (possiblibly outdated)
> understanding of what machine learning models can and cannot do with
> regards to their training data, and a reading of parts of the GPL 2
> and 3, suggest that at best the SWH's LLM is in a legal grey area and
> at worst directly violates the license of GPL code that it ingests for
> training. As such, I don't think it is accurate to say "you cannot
> prevent people to use “your” free software for any purposes you
> dislike" in response to concerns about automatic inclusion of free
> software into LLM training sets. Specifically, my understanding (as of
> a few years ago) is that LLMs have difficulty tracing and atttributing
> various aspects of its training to specific inputs, which seems to be
> in violation of of e.g. Sections 5 and 6 of the GPL. Specific quotes
> from those sections https://www.gnu.org/licenses/gpl-3.0.html:

I think that the larger point here is that you do not get to choose who
use your software and for what purpose.  That is the double edges sword
of free software.

Putting aside LLM for a moment, what if some package in Guix is used for
military purpose?  Will this software be removed from Guix because one
of its user uses it in some unethical way, even though it is also used
in an ethical way by others.  Will we penalized users for the sake of
moral high ground?

This raise the question, what is considered ethical and when does ethic
become political dogma?

[...]

-- 
Olivier Dion
oldiob.ca