On Sun, 16 Nov 2025 at 09:21, Francesco Bonazzi <[email protected]> wrote: > > If LLMs are used the right way, they may help with being more productive. > Unfortunately, my fear is that many developers will start using them without > proper supervision. Average number of lines of code written will very likely > increase, but so will the number of bugs.
I think that for a novice SymPy contributor the only right way to use LLMs can be something like to ask questions about the codebase. Using them to write the code just means skipping the thinking process that would be a prerequisite for being able to supervise the LLMs in writing and checking the code properly. So far I have avoided pointing at individual pull requests in this discussion but this one jumped out at me just now: https://github.com/sympy/sympy/pull/28681 It was opened 6 hours ago by an entirely new contributor and in under an hour grew to 800 lines of new code. The author of the PR has also opened a PR in their own repo using the same branch and over there you can see it being reviewed by coderabbit: https://github.com/Cprakhar/sympy/pull/1 There is a comment from coderabbit there that says "Here are the copyable unit test edits:" and then shows hundreds of lines of unit test code that seem to have been copied into the PR. I don't know whether the code in the PR is reasonable. It looks very LLM-style verbose/duplicative but besides that I don't want to review it in any detail if the author hasn't spent time doing that themselves. 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/CAHVvXxQsQwr%3DNujJvEDHn%3DVmmvKAVKmgymOrBGno43V-P_BTgg%40mail.gmail.com.
