Have you considered that the person submitting the PR to fix said problem has themselves experienced it? I find that is most often the case. But there is definitely an increase in the "performance edge cases, or hypotheticals" framing, where I am equally befuddled/too tired.
I guess also AI managed canary l releases could be an option IMHO that combine open PRs.
I came from the "free software as in free puppy" camp. I would stop by to play with the puppy, but I won't feed it.
I guess if I do the same today, I might submit a better-than-band-aid fix from LLM. You would probably think I am not a real user.
I had multiple PRs for entirely hallucinated problems.
Or where no human ever looked at project or PR and were just attacking open source projects with spam PRs pretending to be contributions. Not sure whether it was out of malice or done for some other reason. Maybe they were mistakenly thinking it is helpful?
I believe a fair bit of this is an attempt to gain numbers where quantifiable numbers are advertised on otherwise indistinguishable resumes (with the corresponding belief that those numbers make differences in filtering or hiring decisions).
I base that on having seen people in the days of old put their Stack Overflow reputation on resumes thinking that having a rep of 4321 will be more likely hired than someone with a rep of 1234.
Having "210 accepted PRs on GitHub" (they believe) would appear to make them a more productive / less risky hire than someone who hasn't done any.
---
This is really a case of indistinguishable resumes where nearly everyone in the pool went to the same college, same internships, same capstone project, and maybe the same first year or two of professional experience.
I could see it as a "proof of familiarity with CI and expectations for a PR", though for applicants that are spamming the process with poor PRs (and people posting poor Q&A on Stack Overflow back in the day), this often hurts them more than it helps as its a demonstration of what they are going to be doing if they are hired (I've already got too many poor quality PRs to review, your PRs would not reduce the amount of time that the reviewer would need to spend).
So yeah, I've considered it and eliminated it for most PRs we used to get. Because no, a 2nd year student in university did not run into an obscure edge case in production that none of my 1M+ QPS users did.
For me they're real issues. Half of which i'd found before but never had the time to submit something for, the other half its things the AI will have encountered - and i check to confirm its a real issue.
The AI will follow guidance you leave in a repository. One of them said "no ai", so it reported that to me and I just forked and went on with my life.
There might be people farming for gh contribution i dontk now — but I can assure you the trend will continue, and afaict it is improving software; at least most of my contributions have been merged.
I see a lot of unsolicited PRs (and my case not from actual users but instead university students trying to pad their resume with random open source contributions but while clearly using AI exclusively)
AI can help with that! Give an agent some rules, like via a skill for the project, and then point it at the PR/Issue for triage to determine if it's one that fits the rules for auto-closing (already reported and decided WONTFIX, for example). Have it run this across all your repos.
Have the AI tooling do the heavy lifting of reproduction and benchmarking. One of my projects is fairly complicated and one user ran into some problems which he could describe the situation that caused it, but was light on any other details: my apt package cache was slow to serve the PHP update to his users. I asked AI to try to reproduce it: Set up a repo, set up different versions of my cache, have a client pull it, put an updated package in there, and have a client pull it again under different configurations. This could have taken me hours to do myself. AI did it with ~2 minutes of my effort.
>I have not organized my life around triaging annoyances
Then don't. AI can help with the triaging. Or, I give you permission: Close the repo for contributions. If you don't have time to maintain it, let people know, so they can fork or replace it. It sounds like your repos are already closed to contributions, due to lack of time/attention. That's fine, don't feel bad, just go ahead and make it official.
Not all contributions are equally valuable, and that is notwithstanding AI use. AI facilitates volume but not necessarily quality. Why should anyone spend resources benchmarking and reproducing a non-issue, or a PR that will not otherwise pass muster? We can automate a lot, but human judgment is still critical in making things to be used by humans.
Sure; if the primary repo goes unmaintained, and nobody else steps up to maintain a fork, yes you might get multiple forks that solve a specific users individual issue. I'm not sure there's a way around that.
The comment I was replying to sounded like someone who was unavailable to handle non-AI requests as well. AI contributions is making it worse, though my biggest projects haven't been hit by it yet <knock on wood>.
Low quality contributions is a very real problem, I agree with that.
You are too nice. Anyone who isn’t going to spend time to explain an issue should be automatically rejected.
We’re in this ugly situation where AI causes the problem and people’s idea to fix it is throw in more AI.
Just like throwing technology at climate change when our own obvious habits are destroying the planet. Let’s move to Mars.
I like the idea of leaning on a decision like Jev model to identify such things.
Table stakes for being an engineer these days should include knowing how to control LLMs, it's as fundamental as a working knowledge of databases. One doesn't need to know how design a data model from scratch or code up that python script using the OpenAI API from memory, but should know the general shape of what's possible and how to look it up.
For me it was a bit useful but it has not offset even 10% of damage caused by LLMs
and is solving least important problem
This requires a huge setup of the necessary AI infrastructure.
Free software and open source licensing, and projects like Debian, are set up so the software survives even the death of the author and the project. They can make their own improvements for their own ecosystem without upstream support. They integrate what comes from upstream whenever they can.
Someone getting their contributions accepted by upstream is a nice thing to have, but upstream don't have to accept it, and if they do, it's at their pace.
Secondly, don't let a pile of open issues and PRs bother you, unless they're fundamentally incorrect (in which case let your authorial voice be known). Because it's also useful for users of your software to give each other feedback and help each other out, e.g. "this app doesn't work with the latest macOS changes, but here's the code changes you'd need to fix it" means that users can help each other often for years before the maintainer gets around to looking at issues. If you're on some sort of code forge site, users can organise there. If you're not (and only accept patches by email or whatever), users can organise themselves elsewhere.
This is what I do, and have done, since long before LLMs. And if they aren't interested? If its a big enough problem for me, I maintain my own fork. Forking is the greatest part about open source projects!
And be OK with closing issues that have dubious merit?
> Let me be the agent of a trove of people telling you what to do and how to do it
Do you see the problem here?
I rolled my eyes so hard I died.
That being said, I know if one reaches that point in burnout, money won’t change anything. You are under no obligation to do any of this after all. So in the meantime, thank you. :)
We need the equivalent of that for open source software maintained by people from before 2004.
This would eliminate a lot of AI slop, and I wish I had a popular repo so I could try it. I might unfilter and glance at the non-funded PR's once a year.
As a maintainer of some projects: at this point I would appreciate far more useful review of already submitted PRs.
As long it is not blind copy-pasting LLM output. That would be highly obnoxious attack.