On 2026-07-29 23:22, Daniel Colascione wrote:
An LLM would be the perfect tool for making changelog commits, BTW. I
*hate* the things and over the years, I've kept a change local instead
of bothering to itemize in a commit message all the changes git already
records that I've done.
I agree on that, as I use locally running and free-software LLMs to
generate descriptions of website pages, YouTube videos, etc. and it
gives me great value, and similar is for commit messages.
Emacs is the tool for thinking and writing, its developers are some of
the best programmers out there. Why ask Claude, of all things, made by
Anthropic? It's not just a policy slip but it goes against everything
free software stands for.
This idea that LLMs are crutch for poor developers isn't something you
observe in the broader software world. You can hate LLMs all you want,
but more is happening than slop. The Linux kernel derives tremendous
value from LLM code review. Are Linux developers any less skilled than
Emacs ones?
I don't know which information you use to think I hate LLMs. I am
running it all time on my computer. Though in GNU project we strive to
foster free software, and not to give examples how to use proprietary
LLMs, see:
Who Does That Server Really Serve? - GNU Project - Free Software
Foundation
https://www.gnu.org/philosophy/who-does-that-server-really-serve.html
There is no rule that any free software user should not use proprietary
software. Yet we should not be mentioning on mailing list or pointing to
non-free or Service as a Software Substitute, or SaaSS.
It's hard to watch the anti-AI sentiment brew here.
I can't see what you see. I do not see anti-LLM sentiment brew here. And
I don't see any AI yet.
I see GNU project and FSF and developers being aware of copyright
implications and in the process of getting conclusive statements and
policy on that.
An individual may not care how to use LLM generated text and so may
adopt the attitude "everything goes".
FSF and GNU project cannot, those are serious issues that impact
possibly millions of people and their lives, and that is why there is
urge to get the final conclusive policy on LLM generated code and
contributions of such.
AI is not evil.
There is now 450 models on my devices. Guess what?! There is few of them
evil. What you speak of "AI", is the language model outputting text or
image by probability. No intelligence there. It can be very evil as it
depends on the model authors.
You can say it's unethical appropriation of work; I say it's a boon to
all mankind, an inversion of the Hobbesian war of all against all: AI
is all teaching all. I see no incompatibility with free software. We'd
do better trying to make models prioritize user freedom than trying to
undo the future. Isn't free software supported to be about empowering
users to control their own computing? An AI can generate personalized
software on demand. What could be more freeing?
It is definitely bringing benefits, yet it is not good conflating
utility with liberty. One thing is using the tool, other is being held
hostage by the infrastructure that sustains it.
If the service is free as in beer but proprietary as in freedom, you are
at the mercy of their Terms of Service. They can change the model, raise
the price, or censor the output. A truly free software approach would
allow you to spin up a local model, modify the weights, and run it
offline. That is control. Using a SaaS LLM is just another form of
"Service as a Software Substitute" (SaaSS).
Ownership: Do you own the output? Do you own the tool? So that is why
legal questions are yet to be answered on that.
I also can't see a universe in which software gets rolled back to the
blissful familiarity of 2022. If LLMs end up conflicting with copyright
law, it's going to be copyright law that gets steamrolled. Do you
really think that the past four years of tech is going to get rolled
back in an orgy of copyright lawsuits? Come on. That's ridiculous.
To get steamrolled is oversimplification.
Bartz v. Anthropic (U.S. District Court, N.D. Cal.) Found that training
AI on books is "fair use," but that Anthropic's storage of pirated books
was unlawful. Led to a $1.5 billion settlement.
https://www.reuters.com/world/us-judge-approves-anthropics-15-billion-settlement-copyright-lawsuit-2026-07-20/?utm_source=ainews247.org&utm_medium=referral&utm_campaign=ai_news_stories&utm_content=story_dashboard_id_262364
GEMA v. OpenAI (Munich I Regional Court, Germany) - The court ruled that
memorization of copyrighted texts within an LLM model constitutes a
reproduction under copyright law, and the TDM exception does not apply
to this -
https://www.osborneclarke.com/insights/gema-vs-openai-ai-memorisation-reproduction-relevant-copyright-law-and-tdm-exception-does
Advance Local Media v. Cohere (U.S. District Court, S.D.N.Y.) - The
court denied Cohere's motion to dismiss, allowing publishers' copyright
infringement claims to proceed to trial. This case demonstrates that the
fair-use defense is not a guaranteed win for AI companies. It allows the
case to move forward and test these arguments in a new context with news
publishers.
https://case-law.vlex.com/vid/advance-local-media-llc-1094877471
Eschewing AI use on copyright grounds, and especially being so paranoid
as to force-push to master to remove a tag, serves as much purpose as
maintaining a stable of horses in case this newfangled automobile
technology gets banned.
The horse/automobile analogy fails because cars didn't ingest every
stable without permission, nor did they raise questions about who owns
the destination.
The $1.5B Anthropic settlement and GEMA v. OpenAI ruling show copyright
law isn't being "steamrolled"; it's actively shaping outcomes.
Force-pushing to remove a tag isn't Luddism but maintaining legal
provenance and the GNU project has always been about choosing tools that
respect freedom and not rejecting progress, but rejecting dependency.
--
Jean Louis
---
via emacs-tangents mailing list
(https://lists.gnu.org/mailman/listinfo/emacs-tangents)
