Hello friends, Hope everyone is healthy and enthusiastic! It's great to be part of this group. I look forward to engaging discussions and sharing ideas with you all.
https://www.titoreista.com On Thursday, November 13, 2025 at 8:36:30 PM UTC+7 Oscar wrote: > On Thu, 13 Nov 2025 at 12:03, Daiki Takahashi <[email protected]> wrote: > > > > Let me make this clear upfront: all of my posts on GitHub, including > this one, rely on translation by an LLM. > > > > I believe it would be reasonable to explicitly state in the policy that > spam-like PRs and PRs relying heavily on LLMs are prohibited. > > It is very difficult to define what is meant by "spam-like" and I > doubt that someone submitting a PR would understand this in the same > way as reviewers would. > > There are different ways of using LLMs and the way that you use them > is absolutely fine. The way that many novice contributors use them is > not useful at all though and at least right now is harmful to sympy > development. I'm not sure how to define the difference between those > in a policy though. > > > Along with that, the policy should also clarify that such PRs may be > proactively closed without prior notice, > > and that there should be a clear process for appealing an incorrect > closure. > > Realistically I think in some cases this is the only option. Just > deciding to close them is still a burden though. > > > To reduce the review burden, one possible approach would be to require > all PRs to undergo an initial review by Copilot before human review. > > However, I am not sure how capable Copilot actually is. > > I don't know about Copilot specifically and actually there are many > things called "copilot". I have used "GitHub Copilot" which is an > editor plugin for autocomplete but now there is a "Copilot" button on > the GitHub website that is something different (more like ChatGPT). > Does anyone have any experience of using that? > > I see better potential in using AI to help out with reviewing PRs than > having people use AI to write the PRs. Many PRs need quite simple > feedback like "this should have tests. Please add a test in file f" > that could easily be handled by AI (and probably in a more patient, > friendly and helpful way than feedback from human reviewers such as > myself). > > Somewhere someone suggested using CodeRabbit which I have seen on some > other repos. I haven't seen it produce anything useful but supposedly > it gets better if you "teach" it. > > -- > 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/cad68c0c-faeb-424a-8ac8-75c5b16ba243n%40googlegroups.com.
