On 2026-05-28 05:02, Lars Noodén via libreplanet-discuss wrote:
On 5/27/26 21:56, Jean Louis wrote:
LLMs are generative synthesis engines that operate on probabilistic
reasoning, contextual understanding, and semantic compression.
No, LLMs have nothing to do with generation. They merely ingest
'tokens' and recombine sets of them into statistically plausible
combinations. Lots of people get confused about that, perhaps (to be
generous) because of wishful thinking on their part.
That is not literally correct. Maybe you mean "creation" like something
new, I agree on that, LLMs do not create, but they do generate. The "G"
in GPT stands for "generative". So yes, LLMs are generative in the
technical sense: they produce new sequences of tokens that didn’t exist
verbatim in training, by sampling from a learned probability
distribution.
Your underlying point is important, you probably mean "no creation"
there. I just think you are denying that they are generating in the
sense of creating meaning or having intent.
Plagiarism requires intent.
Intent is not a prerequisite for plagiarism as defined in academia over
the ages.
LLMs do not have any intent.
The definition you are referring to, refers to human, as only human can
have intent. Not the computer alone without human.
Let us not conflate the definition where using someone's work without
attribution is considered plagiarism with or without intention. That
refers to human. Not to machine.
Agreed, LLMs have no intent. Thus LLMs just plagiarize despite being
automatons. But those who interface with LLMs can have intent. One
can question the intent of those interfacing with LLMs because most
people know, at least on some level, that the 'tokens' are slurped up
verbatim from the WWW at large without attribution.
So let us say, you ask the LLM to give some full work at once, like
doctorate, and then you could maybe get plagiarism verbatim.
But if you ask LLM to construct your sentences, paragraphs, by using
your own thoughts, that isn't plagiarism. You create by using your mind,
the LLM is just a tool.
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.
In the context of code, one recent example is the use of LLMs to try to
strip the Chardet project of its license, among quite a few other
similar attacks some more public some less public. Noticeably, the
original code is in the training set so the stripped version can hardly
be described as a cleanroom implementation.
People can do that, but that deviates from the original statement how
LLMs "main" use is something like that. Obviously it was human, not the
LLM, human who used LLM to generate slightly different copyrighted code.
So don't blame the tool. Same tools could be made with software that is
not LLM. Should we then generalize and call all software as copyright
infringing because someone used software to generate copyrighted code?!
That makes no sense.
It is also insulting to people who run local large language models,
someone comes out of the blue and generalizes, it feels like all of my
friends are plagiarist or something, while reality is totally different,
and such statements come from people who don't even run any local LLM or
are they say they are not experts anyway.
--
Jean Louis
_______________________________________________
libreplanet-discuss mailing list
[email protected]
https://lists.libreplanet.org/mailman/listinfo/libreplanet-discuss
