I think a lot of the issue with these AI submissions isn't so much that there's AI slop code that'll ruin everything. I mean yes, I'm sure there's plenty of that. But a lot of these prominent open source projects were hoping that the people making the submissions actually knew what they were doing and were doing it mostly for the love of the project. Or because they work for a company that is also maintaining it.

The problem I believe, is the motivation behind all of this. I'm going to guess that a lot of the submissions aren't useful because:

• They don't actually fix anything. They may just be rearranging code around
• They attempt to address something that didn't need fixing
• They may have found a legitimate bug... but so did the 100+ other people who did the same thing.

And everyone wants their name on the Linux kernel repo log so they can go show an employer or someone "LOOK, I CONTRIBUTED TO LINUX!!!!11 AREN'T I LEET?"

shuffy_losysI think a lot of the issue with these AI submissions isn't so much that there's AI slop code that'll ruin everything. I mean yes, I'm sure there's plenty of that. But a lot of these prominent open source projects were hoping that the people making the submissions actually knew what they were doing and were doing it mostly for the love of the project. Or because they work for a company that is also maintaining it.

The problem I believe, is the motivation behind all of this. I'm going to guess that a lot of the submissions aren't useful because:

• They don't actually fix anything. They may just be rearranging code around
• They attempt to address something that didn't need fixing
• They may have found a legitimate bug... but so did the 100+ other people who did the same thing.

And everyone wants their name on the Linux kernel repo log so they can go show an employer or someone "LOOK, I CONTRIBUTED TO LINUX!!!!11 AREN'T I LEET?"

You hit a lot of the points, there are a few more though.

The config:

sometimes changing a variable fixes an issue, but it can break it for others. The issue I have seen are people try and compensate by being “less destructive” but really they just show there hand that they don’t fully understand the issue.

A variable change that fixes it for you but breaks for others actually probably requires detection logic before the function not a change to the function itself.

That leads me into the other issue that stems from such.

Maintenance:

You fixed the bug and AI didn’t even mangle it. But you still don’t understand the product, or flow, you might not even be a software dev and are straight vibe coding. That’s not what some maintainers want, they want or maybe even a project NEEDs someone that can help maintain it, and flyby patching even if it works by people that don’t stick around is just more code someone overworked needs to remember and understand for you.

I’m not really against AI or help. I am not even against nobody’s using it to try and help.

But it does cause churn and it can be exhausting.

Please contribute smartly. When in doubt open a draft PR, start a discussion. Use your own branch for your goal and let a maintainer take over your draft when they can.

Edit edit edit: also if for whatever reason you cannot help yourself. Ask your AI to follow the repo convention and try to match stylistically.

Non software folks maybe don’t know, but the robot will try to code by the book usually the languages standard, that’s fine, but anyone who is into coding and has met others knows that coding is like language and you can have accents and speech patterns.

Telling AI to read the repo first and match stylistically can help your code “fit”.

shuffy_losysI think a lot of the issue with these AI submissions isn't so much that there's AI slop code that'll ruin everything. I mean yes, I'm sure there's plenty of that. But a lot of these prominent open source projects were hoping that the people making the submissions actually knew what they were doing and were doing it mostly for the love of the project. Or because they work for a company that is also maintaining it.

The problem I believe, is the motivation behind all of this. I'm going to guess that a lot of the submissions aren't useful because:

• They don't actually fix anything. They may just be rearranging code around
• They attempt to address something that didn't need fixing
• They may have found a legitimate bug... but so did the 100+ other people who did the same thing.

And everyone wants their name on the Linux kernel repo log so they can go show an employer or someone "LOOK, I CONTRIBUTED TO LINUX!!!!11 AREN'T I LEET?"

Information overload is a thing.

on the implications of AI and how it destroys communities (not by me):
blog.fallible.net/it-breaks-a-village/

they should make a whitelist of selected maintainers who can submit manually reviwed llm generated code. anyone else should be banned from using it.

CaptainDodothey should make a whitelist of selected maintainers who can submit manually reviwed llm generated code. anyone else should be banned from using it.

If people were honest enough to always mark their work as vibed, we wouldn't be having this problem in the first place. But these spammers are already trying to game the system by pushing the slop, and the machines that produce said slop are being intentionally optimised to make telling them apart from humans even more difficult.
Noy to mention that the harm these tools bring isn't emerging from scaling (like, e.g., the internal combustion engine wrt polution), it's its multiplying effect of what harm a few numbers of sources can do.
God knows how many malicious individuals are running reputation/karma farming campaigns right now... :shadedshu:

It should be noted for those that didn’t look though, that he is not “stepping down” as in. “Not doing it.”

He is just not reviewing commits to the broader project and only doing ones in the parts he maintains.

Again:

I unfortunately lack the bandwidth to generally be able to offer review except in those areas I maintain.

I have separated the patches into areas maintained by David, Andrew, and Alice + Danilo, so each can take the patches relevant to the subsystems, for convenience.

Little sensationalized but that’s TPU now.

Solaris17It should be noted for those that didn’t look though, that he is not “stepping down” as in. “Not doing it.”

He is just not reviewing commits to the broader project and only doing ones in the parts he maintains.

Again:



Little sensationalized but that’s TPU now.

Exactly, he said he can't handle the amount of PR he has to review and splits the work between other maintainers.

ShihabIf people were honest enough to always mark their work as vibed, we wouldn't be having this problem in the first place. But these spammers are already trying to game the system by pushing the slop, and the machines that produce said slop are being intentionally optimised to make telling them apart from humans even more difficult.
Noy to mention that the harm these tools bring isn't emerging from scaling (like, e.g., the internal combustion engine wrt polution), it's its multiplying effect of what harm a few numbers of sources can do.
God knows how many malicious individuals are running reputation/karma farming campaigns right now... :shadedshu:

I don't think marking work as "AI generated" solves anything really
Not all AI generated code is bad, some is quite good

The problem is submission overload, and to review all this is the thing that burns people out, since before, there wasn't that many people that would do PRs, but now, the floodgates are open and they are getting PRs left and right and it's just so much extra work

There have been quite a bit of slop, and I don't mean the AI generated kind in the linux kernel, and they went through the review process easily enough only later to find out, WTH were they thinking

so it's a pure number issue here, there are only so many maintainers, and they got other things to do as well, besides trying to review 1000's of PRs, and it's just too much work

They change too much which should not be changed

claude, gemini, copilot do not understand bash and shell code. When tehy can not even read the manpage, use wrong assumption, do not understand lvm2, etc. i highly doubt they can do any C or rust code properly.

shell is much easier as C or rust. Claude destroyed data several times this week. I want to learn something new, use something new where of course i also ask those for suggestion. than i test it and see it myself. Fact is, two times this week over 1GB of data every time was lost because of those super intelligence / ai / llm whatever you name those


A german news outled, very badly technically written, claimed, the linuxkernel team dropped 13 year old code. it was only used for one thing and no one really understood what it did anymore.

The ai slop will ruin certain projects in the long run and there is no fix yet.

edit, the gemran outlet which reported this earlier. Yes, that guy only does his own projects anymore because the ai slop sumission is too much to do. he only focus on his code parts and does not review all the stuff he did before

A Computer GuyInformation overload is a thing.

This is really the issue.

AI is a force multiplier at its core. And it gets hard to tell who is using it "well" vs what is absolute garbage when you have a billion submissions on your plate. This was never an issue when there were far fewer submissions, because a human could review them all.

I'm not sure what the answer is, frankly.

_roman_claude, gemini, copilot do not understand bash and shell code. When tehy can not even read the manpage, use wrong assumption, do not understand lvm2, etc. i highly doubt they can do any C or rust code properly.

I've been using claude in a project recently. At least the opus 5.5 model absolutely can write bash shell code, and even complex C programs. It's actually shocking how far they've come. I'm not using them as a "replacement for the human" like techbros dream, but they certainly are more capable than I'd thought.