Hi André,

On 2026-05-28 at 18:12-03:00, André Batista wrote:
> It was the company that developed, trained and decided to offer
> this as a service WITHOUT WARNING THAT WHAT THEY PROVIDE
> CANNOT BE USED FREELY, so they are the ones primarily responsible
> for the infringements they induce others to cause.

Emphasis mine, and that is not the reality.  In the terms of use
of Copilot (which is responsible for cost-ineffective autocompletion
in the most popular editor that nobody here uses):

> we can't promise that any Copilot's Responses won't infringe
> someone else's rights (like their copyrights, trademarks,
> or rights of privacy) or defame them.  You are solely responsible
> if you choose to publish or share Copilot's Responses publicly
> or with any other person.

OpenAI and Fæcebook also admitted to the memorization "bug",
as cited in US copyright office's draft of Copyright and AI,
part 3, section II.D.2.

https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf

On 2026-05-28 at 18:12-03:00, André Batista wrote:
> [A free software project] cannot possibly be held responsible
> for clearing the copyright claims of all works that are offered
> and can be assumed to be working in good faith
> when accepting patches without checking.

I fully agree, there is no reason to not trust a contributor's
claim of distribution rights to a patch.  What I'm arguing here
is if contributor C presents patch P from source S, S says
P may or may not infringe some copyrights, and C says P does not,
despite having no means to verify that claim, it's clear
that C is delulu and not to be believed.

Let's take another example, if C gives us a car P from manufacturer S,
while insisting that it doesn't happen ever despite S saying
P fails to brake 1% of the time, would you trust C and ride P around?
What would any court thinks if you hit someone with P
and blame C's assurance?  Of course, it would be a different scenario
if C withheld the car's model.

On 2026-05-28 at 18:12-03:00, André Batista wrote:
> qui 28 mai 2026 às 17:54:14 (1780001654), Nguyễn Gia Phong enviou:
> > There is practically zero inclusion of LLM output in Guix codebase.
> > Attempts to introduce legal, technical, and moral risks, etc.
> > to the project should preach for acceptance, not the other way around.
>
> Acceptance?  Not sure if I understood this last sentence.

If big tech boosters want to change to status quo and incorporate
LLM outputs along with their risks into free software projects,
they should present their evidence and do the convincing.

I am disappointed that the discussion here treats LLMs as inevitable
and caters to their users instead of securing our software freedom.
If one day all courts worldwide rule that all LLM outputs
are under the public domain, then it's not too late for another GCD
reverting the their ban anyway.

Best wishes,
Phong

Attachment: signature.asc
Description: PGP signature

Reply via email to