I have been following this discussion for a while. What I do not understand is this: why would anybody want to push a PR which he does not understand? This seems to take out all the fun. I
[email protected] schrieb am Mittwoch, 17. Dezember 2025 um 14:28:42 UTC+1: > I agree with Oscar on AI generated PRs. Not only sympy many other open > source projects are experiencing this issue, I see most of them are closed > without merging as they are verbose and often contain false claims about > what they do or haven't passed basic unit tests. I think having a clear > policy on disclosure and understanding of the code is the only way to > manage this moving forward. > > > > On Wed, Dec 17, 2025 at 12:17 AM Oscar Benjamin <[email protected]> > wrote: > >> On Tue, 16 Dec 2025 at 17:48, Arka Saha <[email protected]> wrote: >> > >> > I read this discussion on AI bots performing PRs and indeed it's an >> issue, and would probably be a major issue in the next few months. But >> again, if the code so generated, even by a AI bot, if it performs well, >> benefits the organisation and solves the issue, then I don't see why we >> should demotivate such cases. >> >> It isn't good code though and it doesn't benefit the organisation or >> solve any issues. >> >> What we are seeing is really just spam with many more pull requests of >> much lower general quality. Even if some of them are good, the review >> process is overloaded to separate the good ones from the rest. >> >> The problem is that AI is enabling novices to generate low effort PRs >> much more easily at the same time as making it harder to review those >> PRs because superficially they look good but actually everything about >> them is wrong in ways that are hard to predict or understand without >> close attention. >> >> Because the AI can write the code and open the pull request and answer >> all of the questions about how to do all of those steps, people think >> it is acceptable to do that without having done basic things like: >> >> - Reading any of the code (before or after the changes). >> - Thinking about any of the code or changes themselves. >> - Knowing what changes are even in the PR that they have submitted. >> - Knowing how to test changes to the code or how to run the test suite. >> >> Previously it was not really possible to get to the point of having a >> PR that passes CI checks without spending some time doing these >> things. Now it is possible to skip all of those steps and then produce >> a garbage PR that superficially looks reasonable while actually being >> entirely wrong. >> >> People at this level really are not benefitting from the use of AI. If >> they learned how to do things without AI then they might become >> capable of using the AI to produce something good in future. >> >> -- >> Oscar >> >> -- >> You received this message because you are subscribed to the Google Groups >> "sympy" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> > To view this discussion visit >> https://groups.google.com/d/msgid/sympy/CAHVvXxRzz9ucfNJqT81xw81aNtRkSjywztq0n%2BKVKZJKFkDomA%40mail.gmail.com >> . >> > -- You received this message because you are subscribed to the Google Groups "sympy" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/sympy/a8d3d6c1-d98e-423b-97be-84a0ea8e9d5an%40googlegroups.com.
