On 2026-03-11 08:26:51, Ulrich Müller wrote:
> IMHO there's nothing dubious here. An algorithm by itself (i.e. its
> ideas or conceptual steps) is not copyrightable, so you can take any
> algorithm from TAoCP, as long as the implementation is your own.
>
> On the other hand, a concrete implementation of the algorithm is
> copyrightable, so a verbatim copy from the book would not be allowed.
>
> ...
>
> So if a package would contain any significant amount of material from
> TAoCP then it seems pretty clear that we needed "all-rights-reserved"
> in its LICENSE.
The situation is not so clear to me:
1. How do you know that the code the LLM generated was taken from
TAoCP (or not)?
2. AFAIK there has not yet been a legal ruling on whether or not
plagiarism-washing though an LLM constitutes a new and original
work.
If you see code that you *know* is stolen, then all-rights-reserved
would of course be appropriate. But for the other >99% of what an LLM
spits out... we know it was stolen from somewhere, but we don't know
where, or what license it was under, or if the parent company donated
enough money to make it legal.
Setting LICENSE="dubious $LICENSE" would indicate that there is enough
obvious LLM use that the license cannot be trusted, even if we don't
have a copy of the victims on our bookshelves to point to. The need
for every license in $LICENSE to be in $ACCEPT_LICENSE maps this
nicely onto the UI in my opinion.
