On Mon, May 25, 2026, 10:16 AM E. C. Masloch via Freedos-devel < [email protected]> wrote:
> > > On at 2026-05-25 09:05:04 -0500, Jim Hall via Freedos-devel < > [email protected]> wrote: > >On Mon, May 25, 2026 at 8:16 AM Ben Collver wrote: > >> > >> Just FYI, FreeDOS is already on the Open Slopware list [1] for two > >> merges of AI generated code into COUNTRY.SYS [3] [4]. > >> > >> -Ben > >> > >> [1] > >> https://codeberg.org/small-hack/open-slopware > >> > >> [2] > >> https://github.com/FDOS/country/pull/16 > >> > >> [3] > >> https://github.com/FDOS/country/pull/18 > >> > > > > > >Yes, I know about those. And as I explained to the person who asked > >about it, these aren't showing code written by AI. It's using Copilot > >to analyze. > > > > > >#16 shows Copilot writing a pull request overview. See here: > >https://github.com/FDOS/country/pull/16#pullrequestreview-3738866735 > > > >..and here: > >https://github.com/FDOS/country/pull/16#pullrequestreview-3748854780 > > > >#18 shows that Copilot "wasn't able to review any files in this pull > request." > > > > > >So in both cases, the code was still written by a human, but Copilot > >assisted in a review. > > > >I disagree with their assertion that using Copilot to generate > >comments on a pull request "makes [AI] slop acceptance more probable > >and maintainability more difficult." > > I would also argue that, even if there were contributions to it, > country.sys doesn't contain any code actually, it's to my knowledge purely > a data file. > > The new DR-DOS v9 (which is proprietary and closed source) is likewise > reported to use generative AI only for documentation and tests, not for > generating program code. > > I do think that makes a notable difference. > > Regards, > ecm > I agree, country.sys is strictly a data file, not a program itself so can't by definition be ai generated "slop" code. Honestly that sites reads horribly and it seems like ai slop itself given it didn't seem to even read what it points to as references. Both of those were just me having some fun and seeing what a new GitHub feature does and if its even worth using. I was treating it like any other tool for static analysis and continue to believe any and all tools that can help find potential problems in code are worth evaluating. The one place it offered suggestions was in the Python code that is ran as part of the CI checks and not for country itself. >From my background, both my undergraduate and graduate work involved "ai" specifically artificial neural networks, so modern LLMs are a curiosity for me and I enjoy playing with them to see what they can do. So far I have no confidence in their programming abilities beyond advanced template implementations and good fuzzy search and replace tools. But they are improving and can aid in researching (pointers to references and summarization, not the information itself which is too often synthesized [made up]). >From my reading that site's classification simply means someone who is not opposed to using ai tools where they make sense and not the as implied by using the word slop, meaning of publishing ai generated garbage, so a totally useless classification. It's like saying I don't like software that uses a linter or automatic formatting tool so I'm putting them on my naughty list. I agree people shouldn't use tools and their output without understanding, but just like any tool, its the person using it not the tool that one should be concerned with. Anyway, no actual code has been ai written, and even if permissible, I wouldn't publish code I didn't largely write, rewrite, review, and understand. As for the policy itself, consider exceptions for translations, generating boilerplate like tests for ci, static analysis for finding issues, and perhaps ci infrastructure tooling. I am fine with no ai generated code in main DOS programs, as I don't see it helping much there anyway. Jeremy
_______________________________________________ Freedos-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freedos-devel
