Hi,
I haven't contributed code to the project, so treat my opinion accordingly if
that's important.
I think that all code should be scrutinized equally. Bad code has been produced
by both people and LLMs. So the most important thing is code review because,
ultimately, bugs, regressions, security and performance issues can be
introduced regardless of how the new code has come into existence.
One case I dislike about LLM-based contributions is non-trivial and/or large
code submissions when the contributors don't even understand the code. That
creates the ridiculous situation where the reviewer(s) can't get answers from
the submitters, about why something has been done one way or another.
So it makes a lot of sense that the contributions should be 100% understood by
the people submitting them. Whether an LLM was used in the process, I'm more
relaxed on that if it's about formatting or the "grunt work" of restructuring
some data that can be easily validated programmatically.
It's certainly easier to review LLM-generated contributions, since LLMs have no
feelings, while people do, but bad code should not be merged either way. Just
the same, I think that good code should not be thrown away just because an LLM
was used in the process - as long as it's well understood, it passes the review
for code quality and it does what it's supposed to do without negative side
effects.
--Nick
On Sunday, May 24, 2026 at 06:31:41 PM EDT, Jim Hall via Freedos-devel
<[email protected]> wrote:
Hi everyone
Like many of you, I'm concerned about AI creeping into open source projects.
I've added a few other issues at the end [1] that expand on this.
So there are reasons not to allow AI in the core parts of FreeDOS. I thought
I'd try to capture the consensus of the email list into an AI policy that I can
post on the website.
I've tried to keep this high level. I think this matches the conversations
we've had here, whenever AI has come up:
Scope: Any package that gets installed as part of a “plain DOS” installation.
Packages in Games or Devel or Edit or Util (out others) are not “core” to
FreeDOS. I'm less sensitive to stuff in the other package groups.
But I'm being careful with my wording here. It's not just Base, but any program
that gets included in a “plain DOS” install. That's where I draw a solid line.
Code: Do not allow AI for code generation.
It's ok to use AI to summarize a contribution (like analyze a PR), or to use AI
to help identify bugs. But the code must be 100% written by a human.
Again, this is only for programs in a “plain DOS” install. If you want to use
AI to “vibe code” a game, or a new text editor or word processor, or something
else, that's up to you.
Documentation: Not sure where to draw the line.
I don't want to read AI-generated bs, I have to do that as part of my day job
(university) and I don't like it.
But I know not everyone is comfortable with writing docs. You might not be good
at grammar or spelling, or even the writing process itself.
Some might prefer to use AI “assisted” tools like Grammarly that can rewrite
sections of text to meet a target, or Scribe that can “watch” what you do and
write a how-to for you.
For me, I prefer to read human-written stuff-- but does it really “break” a
program if AI helped write the docs?
In the end, it should be arguably “written by a human” but “AI assisted” for
docs is ok for me.
If it's your docs, you “own” what's there. If it's wrong, you need to fix it.
If something is plagiarized, you need to take it out.
Or should we not allow AI-written docs at all, for programs in a “plain DOS”
install?
Translation: AI can be used for translating spoken languages.
This has been done for years (and not really “AI” but “maching learning”) such
as Google Translate. While “machine” translation isn't perfect, it can usually
be “good enough” until someone can provide a human-generated translation.
Note that quality can vary when translating to/from different languages. For
example, I find translations from Spanish to English, or French to English, are
usually quite good. Translations from German to English, or Russian to English,
can be pretty rough.
This isn't “translating between programming languages.” See ‘Code’ above.
Did I miss anything?
__
[1] I'm sure AI is a neat tool for some, but using AI in an open source project
can turn off a lot of contributors. For myself, I work on open source projects
because it's fun, and AI is not “fun” for me.
There's also a very strong legal reason; US courts find that AI generated
content is not eligible for copyright protection.
I've also seen examples where AI “vibe coding” regurgitated some of my code.
That was a personal experiment for a “niche” topic where I asked copilot to
write a version of nroff; the AI cited an article I wrote in its summary about
“here's how this works” and I recognized my own code in the generated output.
This is dangerous because in the online article it pulled from, I wrote about
how this isn't a safe way to write this (doesn't catch edge cases, etc) but it
was an easy way to show it to non-programmers. But copilot didn't add that
caveat to the code it generated for me.
_______________________________________________
Freedos-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freedos-devel
_______________________________________________
Freedos-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freedos-devel
