On 2026-05-26 16:28, Lars Noodén via libreplanet-discuss wrote:
On 5/26/26 00:57, Andy Tai wrote:
The US Copyright Office declared that LLM output is
non-copyrightable.
Is LLM output public domain? Then incorporating LLM output into
copylefted program seems safe, as such output is GPL'd.
But there seems to be fear that if much of a program is output
from LLM
(say via gradual patch incorporation or third-party contribution),
then
the copyright of the whole program—GPL or some other license—may
no
longer apply,
I'm curious if such fear is justified.
LLMs are not much more than plagiarism engines.
That LLMs are "not much more than plagiarism engines" is gross
oversimplification that reveals narrow understanding of how large
language models actually work. LLMs are generative synthesis engines
that operate on probabilistic reasoning, contextual understanding, and
semantic compression. Plagiarism requires intent. LLMs do not have any
intent.
If you ask an LLM to explain quantum entanglement in simple terms, it
doesn’t ‘pull’ a definition from Wikipedia. It constructs a novel
explanation using its understanding of the underlying concepts. That’s
synthesis, not plagiarism.
LLMs are trained on vast corpora of human text. But so are humans. When
you read a book and then write an essay inspired by it, are you a
‘plagiarist’? No—you’re learning from context. LLMs do the same, but at
scale. The difference is mechanism, not essence.
You can use any LLM to generate what you wish to call "plagiarism", and
you can use it to be very creative.
How do you generate plagiarism by your definition? Ask the LLM to give
you literal output of some works. Set the temperature to zero. that
specific mode, the LLM acts as a high-speed search-and-retrieve engine,
reproducing existing text with minimal variation. It is citation on
steroids.
Conversely, "authorship" emerges when you use the LLM as an augmentative
layer for your own original ideas. You define the core concept, the
intent, and the direction. The LLM then refines, expands, and polishes
your voice, not its own. It is not generating content from scratch; it
is elevating your expression. The difference is not in the tool, but in
the source of the initial spark.
You’re still mistaking the map for the territory, rather than
discovering the territory itself. Try it out. Method is there, in
previous paragraph.
Thus, in the matter of software, one of the main uses of LLMs is to
strip both attribution and licensing information from whole code bases.
So while it is correct that machine generated *output* cannot be
copyrighted, if it were actually generated from scratch by the LLM.
But, one has to raise questions about the input which it is
regurgitating minus attribution and licensing information. LLM output
is not generated from scratch, it is instead generated from models
trained on licensed code under copyright protection.
There are many LLMs today that are not trained on copyrighted code. We
talk of millions of published models, like those on huggingface.co
I have downloaded 902 models. And I run those which are not trained on
copyrighted sets and which are not proprietary. Now you come with such
mature looking statement how the main use of the LLM is to strip both
attribution and licensing information. Surely you can do that, but it is
not main use. For these models, the 'stripping of attribution' narrative
falls apart entirely.
Yet, you present your statement as if it’s a universal truth. This kind
of broad generalization is dangerous. It’s not just an academic error;
it misleads people who don’t have the time or expertise to dig into the
nuances of these diverse models. They’ll just jump on the bandwagon,
protesting a problem that isn’t actually universal.
Put more effort in research. It is enough to research the website HF I
gave you.
So if you wish to use it in the mode of plagiarism, of course, you are
free to do so. Same thing with actual books, you could take them and
kind of re-write them, without referencing.
If you wish to use it in full authorship mode, you can, there will be no
plagiarism, and LLM is your augmentation tool.
As for software freedom, there is an additional pair of problems caused
by LLMs: using the plagiarized output separates potential project
collaborators from the very projects which are getting exploited, while
at the same time the projects which are getting exploited are getting
isolated from any potential collaborators trapped behind LLMs.
That is under condition that some user use it as plagiarized output. It
simply is not so in majority of cases. I am sure you need to do more
experience with it. Basically, run the LLM.
If something is plagiarized it means user knew it, and had intent to
plagiarize, and would not provide references.
So try doing that, practically, then come back with the report. Then we
have the fact to talk about.
But, yeah, using LLMs to strip the GPL and other software licenses from
code, as well as stripping copyright attribution, is a real problem
now.
For who? Do you have specific case? You can complain. But let us not
generalize.
The abundance of software created with the LLM also generates more
GPL/Apache/Mit/free software. The abundance will soon make the GPL
obsolete, because everybody is then able to generate their software,
which is in the end free software. It is evolution.
There was free software before the GPL.
GPL has the purpose to protect free software to remain free software.
But once you have free software generated on every corner, every
monitor, every computer, the GPL may become obsolete, but users may
still have the freedom.
So keep working on the freedom, don't judge the temporary momentum of
LLm-affairs which you still did not research well.
--
Jean Louis
_______________________________________________
libreplanet-discuss mailing list
[email protected]
https://lists.libreplanet.org/mailman/listinfo/libreplanet-discuss
