Ironically this seems like a perfect use for AI from the maintaner side. "Find low-effort/AI-like PR's without associated issues from new contributors. Discuss with the contributors in the PR about how contributions should be made, then reject their PR's. If any bad behavior is detected in this interaction, block the accounts for 24 hours and add them to a list of accounts to review for permanent banning from the project".
Automated PR's can have automated responses. You make a human effort, you get a human response.
We do similar in Homebrew by autoclosing opens without the issue template (which implies using the API which often now implies AI).
The weird flip side is that the average good AI contribution is better than the average “I used no AI” contribution. Perhaps it’s because Homebrew has so many declarative guardrails and is so easy for agents to test.
I also maintain a popular open source repo and have seen similar. If I receive a low effort single-pass obviously-Clauded PR (many people (agents) don’t even bother using the AGENTS.md file I provide right in the repo?!) with no associated issue I also have no qualms closing. I do generally leave a short note about why I closed (volume of these is around 5/week) showing how far off they (Claude) was and encourage opening an issue where we can start a discussion which rarely happens for the drive-by folks. So I totally feel this but I find myself more disheartened by the practice than angry. It feels like slowly watching open-source go the way of email. Open and free until no-cost spam ruined the inbox for everyone...
> If I receive a low effort single-pass obviously-Clauded PR (many people (agents) don’t even bother using the AGENTS.md file I provide right in the repo?!)
Claude doesn't read AGENTS.md automatically, it only looks for CLAUDE.md.
So the fixes are still fixes, but we (I am also a OSS maintainer) are unwilling to accept them as they boost the contributor’s status where we think the merit is very or extremely limited.
Why not have these PRs counted differently (by the platform), and/or colored differently in the timeline(s) thus made less visible or more clear?
Unless the change is an obvious improvement, it has to be worth the time for the maintainers to spend attention reviewing it (and supporting the code forever, and all the rest).
Even if these particular changes are "harmless" and easy to review, accepting them sets a precedent that encourages an unsustainable flood of AI-generated changes that will overwhelm the project.
These fixes, regularly destroy the architecture, accrue bloat for little gain, refuse to rewrite while demanding to rewrite- and many other such funny noises. Most code contribution by LLMs is garbage if you long-term care about the project. Look at closed source projects that ingest all this madness - windows with its seconds to open the explorer and other catastrophes.
I think Goodhart's law applies really well here. When people make commits for the sake of having made a commit to some famous repo, the "merit" that a commit measures is already useless. Remember Hacktoberfest 2020, before LLMs were popular? https://news.ycombinator.com/item?id=24643894
Because every merge request and every merge comes with cost and (potential) more cost in the future.
Linux lately received a lot of fixes for drivers nobody has cared about for ages. They decided to remove them. Maybe they should have done this earlier, I don't know, but that happened because every code change comes with cost down the line. For real contributers, this cost is much more limited and they have a higher probability of burden the cost down the line.
Pure AI MRs are just rude, because it's just "here take this, don't really care what this is, but AI said it being good, I won't be around, so you will have to deal with this code after the merge and you better don't overlook anything, because I personally don't really know what I am doing, I only know how to have AI making something that looks convincing. Good luck fixing the bugs this change introduces".
My solution is to ask them for proof of work. Many of these disappear after opening a PR that they never tested. How can you claim to have fixed a visual issue without a screenshot? Insta-close.
I think what you're saying is fair, tbh. Imagine you put in months or years into a project to make it a quality piece of work. It develops into something notable and you took all the risk. Then someone comes along to fix a spelling error with a pull request so their name effectively appears on the repo as a "contributor." And you just know right after its going on their resume as "contributed to [...]" or maybe if they're bold "software engineer working on [...]" which implies substantial investment. Then you're effectively sharing credit for YOUR work with someone who did nothing. That is rage inducing. ((Of course: it probably is just juniors trying their best in this horrible industry.))
On the other hand: lets be careful not to dismiss valid but inexperienced attempts to contribute. Having someone want to genuinely contribute to your software is incredibly generous. If someone seems like they're trying its better to give advice than act like a snob because its not good enough. Often pull requests only need small fixes to get in, anyway.
Hi Neil, fun to see you on HN. I agree with your points and I you summarized it very well as "Ultimately, open source is built on trust".
AI is destroying trust in open source and many other areas and I think this will discourage teams from publishing their source code in the future.
On the other hand, personal connections are becoming even more important, which is unfair to the younger generation and people who don't live near tech hubs.
Today's conception of open source is different from the original anyway. It used to be "we made this thing, here's the code" but now it's turned into some kind of weird parasocial thing more akin to Facebook.
I wonder if LLMs can tell when you're taking the mickey out of them. Has anybody tried imitating its mannerisms and had it react? There's a new benchmark for you: SelfAwareness!
Both are fine, but I also prefer 'furnish', as in put in the main parts of your CV. In a way, 'burnish' is more like 'polish', suggesting an extra layer on top, rather than the fundamental parts.
Seems like a contributor reputation score (shared across all projects) that one can build up over time would be a clean solution to the low quality pull requests.
Do recruiters actually care about your open source contributions? Especially now when only LLMs read CVs and match them against strict criteria, I doubt there are many companies that actually care about your work outside of work, in fact they might care in opposite direction - thinking that you'll be distracted from work
Just speculating (I use zero automation when recruiting), but automated candidate ranking systems could well still look for signals that were valuable years ago, but no longer are. It's not a given that people using this stuff even understand the criteria under the hood.
Same story as SEO, I presume. Google surely understood their ranking criteria deeply, yet couldn't exactly win that fight.
This makes sense. Apart from the original thing, I no longer upstream anything. It just takes comparatively more effort than maintaining a private fork with my own idiosyncratic fix. This must mean that even small fixes that non-contributors would make in the past are just happening off the books, so to speak. e.g. I upstreamed a tiny fix to Thrift codegen because some function could be much faster. These days I wouldn't bother. I'd just fork and leave upstream to be upstream.
I suspect many people are like me since it feels like a very normie position to take. That means that contributors are even more likely to be useful because both the drive-by genuine contributors have chosen otherwise and these contributors have increased.
It must be quite painful to be an open source maintainer right now. One wonders what to make of such projects in a future where a feature-list is a sufficient prompt.
Discussing the issues, achieving consensus, making sure the fix fits the existing patterns and actually warrants the change, etc. These are and have always been a challenging part of contributing to open source. Typing code in your private fork that satisfies your “idiosyncratic fix” has always existed and been always the easier aspect of such changes. Chances are many of these changes would never be accepted into the repository anyway in the past.
I ran through exactly this thought process & LLM based resolution just 8 hours ago with a segfault compiling a project which likely doesn't care about the given platform combo much.
GitHub flavored FOSS I believe works best for "corporate FOSS" projects and terrible for "FOSS as how it was conceived decades ago".
Or rather I believe that it is engineered exactly for the former.
What it does well is coordination between corps, working in public (while on a corp payroll) and extracting some drive-by-PR value out of random third-parties.
The closer your project is to that shape, the better it works for you. The further it is away, the more pain you will feel.
This has nothing to do with AI slop.
AI slop just turned up a few knobs that were already there.
Vibe slop or put on your C.V "contributor to <wellknownproject>"? In terms of social currency B wins easily, unless you can guerilla market as well as Openclaw.
that won't work because all those millions of 'projects' are unmaintainable vaporware which is exactly why people try to piggyback on the reputation of software that is actually being used and developed by people
I very much can relate, as I've also been receiving many of these drive-by PRs lately.
One big problem I have with them is that they take away time from project maintainers for reviewing and helping to get the PRs into shape, which then can't be spent on other, more important things. I feel like the "good first issue" GitHub label is specifically attracting these kinds of contributions.
It's not a black-or-white thing though, and you need to tell apart folks who produce slop PRs against any arbitrary repo, from folks using AI to contribute in a sensible way. We've tried to codify some rules in our contribution guide [1]:
- PRs from apparent bot accounts are closed
- PRs from users who file large numbers against random repos are closed
- You're welcome to use AI, but you need to stand behind your PR and be able to explain it
I'm sure we'll adjust those rules over time, but since we have instantiated them, it definitely has become easier to deal with AI PRs and handle them in an a relatively objective way. It absolutely means that sometimes a PR will be closed which could have been an improvement, but I think this is the right thing to do given the circumstances.
I help maintain a big project as well and I can totally relates what the author is describing.
The amount of security advisories and PRs opened is getting out of hands.
This is no longer just spell check PRs that you can merge in less than 1 minute of review. Often that's going to be supposed perf improvement, attempts at fixing things that are not broken, or submitting completely new features.
These relatively low effort PRs takes a lot of effort to review or even just to filter out.
AI code review is nice to have an additional pair of eyes but it doesn't substitute to the maintainer eyes.
A big part of a review is deciding if you want that change merged in. Not because of the immediate code but what it means to the project to bring this in.
AI code review, to my eyes, just reduce the number of bugs, they do not shorten the review time. Otherwise, it means you are compromising on the direction of the project.
> The changes were harmless and correct, but that did not make me feel better about accepting or merging them.
So do you have the project’s best interest at heart or not? If you’re more concerned about the intent of a valid contribution than the content, why don’t you ask yourself where your intent is? You rejected a valid contribution based on unverified vibes about the person’s intent, instead of assuming they were just being helpful.
Why not assume someone does have the best interest of their own project at heart? How do you get to question that out of the gate, while pretending automated grammar and spelling fixes to rack up PR counts are benevolent unless proven otherwise? Simply assume the person who made the thing you didn't make, who was the only person (or persons) on the planet to come up with that exact thing, knows what's best for that person (or persons). Just like you would handle your own stuff.
I think it’s implied that accepting these kinds of PRs will take time away from working on actual improvements to the codebase. So, on the long term it’s better to be strict with these kinds of silly contributions, and focus on changes that matter more (e.g. features or fixes).
Automated PR's can have automated responses. You make a human effort, you get a human response.
The weird flip side is that the average good AI contribution is better than the average “I used no AI” contribution. Perhaps it’s because Homebrew has so many declarative guardrails and is so easy for agents to test.
Claude doesn't read AGENTS.md automatically, it only looks for CLAUDE.md.
Sounds like an opportunity for GitHub/others to better filter these.
Why not have these PRs counted differently (by the platform), and/or colored differently in the timeline(s) thus made less visible or more clear?
Unless the change is an obvious improvement, it has to be worth the time for the maintainers to spend attention reviewing it (and supporting the code forever, and all the rest).
Even if these particular changes are "harmless" and easy to review, accepting them sets a precedent that encourages an unsustainable flood of AI-generated changes that will overwhelm the project.
True, it is actually negative sum, because fake merit erodes trust in real merit, as it makes it a lot harder to spot the latter.
Linux lately received a lot of fixes for drivers nobody has cared about for ages. They decided to remove them. Maybe they should have done this earlier, I don't know, but that happened because every code change comes with cost down the line. For real contributers, this cost is much more limited and they have a higher probability of burden the cost down the line.
Pure AI MRs are just rude, because it's just "here take this, don't really care what this is, but AI said it being good, I won't be around, so you will have to deal with this code after the merge and you better don't overlook anything, because I personally don't really know what I am doing, I only know how to have AI making something that looks convincing. Good luck fixing the bugs this change introduces".
How do you distinguish those who know what they are doing from those who only can write a prompt and copy the result?
Making the haystack bigger is a bad idea if you search the needle
I don't think that there is a technical solution to be found here, as the problem is anything but technical.
__
A hack/trap:
Comment "Ah yes thanks a lot for the hint :)", then make the changes yourself.
Then see how the person reacts to that.
Hack the grifters. Hack the planet.
On the other hand: lets be careful not to dismiss valid but inexperienced attempts to contribute. Having someone want to genuinely contribute to your software is incredibly generous. If someone seems like they're trying its better to give advice than act like a snob because its not good enough. Often pull requests only need small fixes to get in, anyway.
AI is destroying trust in open source and many other areas and I think this will discourage teams from publishing their source code in the future.
On the other hand, personal connections are becoming even more important, which is unfair to the younger generation and people who don't live near tech hubs.
https://www.reddit.com/r/ClaudeAI/comments/1vrlrud/gaslighti...
So funny. And yes, we should 100% find a way to turn this into a benchmark.
Same story as SEO, I presume. Google surely understood their ranking criteria deeply, yet couldn't exactly win that fight.
https://en.wikipedia.org/wiki/The_Sorcerer%27s_Apprentice
I suspect many people are like me since it feels like a very normie position to take. That means that contributors are even more likely to be useful because both the drive-by genuine contributors have chosen otherwise and these contributors have increased.
It must be quite painful to be an open source maintainer right now. One wonders what to make of such projects in a future where a feature-list is a sufficient prompt.
What it does well is coordination between corps, working in public (while on a corp payroll) and extracting some drive-by-PR value out of random third-parties.
The closer your project is to that shape, the better it works for you. The further it is away, the more pain you will feel.
This has nothing to do with AI slop. AI slop just turned up a few knobs that were already there.
Obviously the best way to use AI to furnish your CV is to build your own project using AI.
One big problem I have with them is that they take away time from project maintainers for reviewing and helping to get the PRs into shape, which then can't be spent on other, more important things. I feel like the "good first issue" GitHub label is specifically attracting these kinds of contributions.
It's not a black-or-white thing though, and you need to tell apart folks who produce slop PRs against any arbitrary repo, from folks using AI to contribute in a sensible way. We've tried to codify some rules in our contribution guide [1]:
- PRs from apparent bot accounts are closed - PRs from users who file large numbers against random repos are closed - You're welcome to use AI, but you need to stand behind your PR and be able to explain it
I'm sure we'll adjust those rules over time, but since we have instantiated them, it definitely has become easier to deal with AI PRs and handle them in an a relatively objective way. It absolutely means that sometimes a PR will be closed which could have been an improvement, but I think this is the right thing to do given the circumstances.
[1] https://github.com/hardwood-hq/hardwood/blob/main/CONTRIBUTI...
The amount of security advisories and PRs opened is getting out of hands. This is no longer just spell check PRs that you can merge in less than 1 minute of review. Often that's going to be supposed perf improvement, attempts at fixing things that are not broken, or submitting completely new features.
These relatively low effort PRs takes a lot of effort to review or even just to filter out.
AI code review, to my eyes, just reduce the number of bugs, they do not shorten the review time. Otherwise, it means you are compromising on the direction of the project.
So do you have the project’s best interest at heart or not? If you’re more concerned about the intent of a valid contribution than the content, why don’t you ask yourself where your intent is? You rejected a valid contribution based on unverified vibes about the person’s intent, instead of assuming they were just being helpful.