Conversation

@ljs this feels like the Rust on Linux folks were right all along...?

0
0
0

@ljs the bit where folks resigned, because compatibility measures were blocked by non-owning maintainers that had merge access

0
0
0
@ljs No decisions have been made - it *is* an RFC after all. Expect a lot of discussion, and I would be amazed if this doesn't end up as a prominent topic at the Maintainers Summit in December.
1
0
2
@ljs It *was* sent to linux-kernel.

In any case, it's a docs patch, I'll get my say in things :) More to the point, I don't take policy-relevant docs patches without a pretty clear sense of a consensus behind them.
0
0
2

@ljs Putting aside my general negative thoughts about LLMs, it seems premature to embed this sort of configuration in the kernel. These tools change how they work every couple of months at the minute - it's possible that whatever gets committed will be obsolete very quickly, possibly replaced with something built on Model Context Protocol (MCP), which in turn will probably be obsoleted before the end of the year.

1
0
0

@pbarker indeed. I suppose the argument back would be 'well we can change it then' but I don't like the idea of locking into things like this with the implication that it's ok or this is 'how it's done'.

1
0
0

@pbarker also I think some kind of opt-in by maintainer would have been a good idea.

0
0
0
@ptesarik @ljs @cxberger Hehe. Someone should tell that to kernel CoC people. Apparently taking AI hallucinations, signing them with your and sending that out to confuse people is okay... Signing-off patches based on those hallucinations is okay, too.
0
0
0

@corbet sorry what was that you were saying about RFC?

https://lore.kernel.org/all/20250727195802.2222764-1-sashal@kernel.org/

I think my point has been made here...

1
0
0

@ljs If it gets committed, is it time for a non-AI fork?
Lorenzux?
@corbet

1
0
0
@vbabka @ptesarik @ljs @corbet this will make slop more difficult to recognize, right?
1
0
0

@piggo @ptesarik @ljs @corbet patch 4 includes some rules how usage of "agents" should be marked. I believe we should only focus on that initially + legal/licensing aspects. In fact this series seems to be sent shortly after a different RFC thread by someone else focusing on just that.

1
0
0
@vbabka @ptesarik @ljs @corbet can they adhere to it? Thinking of that post where some moron let an ai agent delete production database by accident
1
0
0

@piggo @ptesarik @ljs @corbet good point, such requirements should not be just in the part of documentation that's supposed to be processed and followed by the agents themselves, there should be clear rules for the people running them, that they should double check.

1
0
0

@vbabka @piggo @ptesarik @corbet to me any series like this is implicitly saying it is fine to use AI tools to write kernel code.

And now you can point automated tools easily at the kernel tree.

Also it is making decisions as to how they are used.

This should very clearly under subsystem maintainer control...

Oh and good luck on relying on any of the rules, see my thread about asking AI about maintainership for a sense of how good this stuff is at even looking up a simple entry in a file

Frankly the way this is being done is ridiculous.

1
0
0

@vbabka @piggo @ptesarik @corbet also we have gone from it will be discussed at maintainers summit 'of course' to it is now heading for merge.

2
0
1

@ljs @vbabka @piggo @corbet
Oh, right, there seems to be some undue pressure! I'm almost getting Jia Tan flashbacks.

1
0
1

@vbabka @piggo @ptesarik @corbet to me we should start with an AI policy document FIRST.

2
0
0

@ljs @vbabka @piggo @corbet Even if we take only the legal aspects into account, what if some jurisdictions decide that AI infringes on other people's copyright? Shall we remove AI-assisted commits again?

1
0
0

@ptesarik @vbabka @piggo @corbet yeah the best way to decide this is for a stable maintainer to just merge this unRFC'd after the last weekend in the cycle.

1
0
0

@ljs @vbabka @piggo @corbet you forgot to add /s.
Just curious: Do I get British understatement right if I say: “I am not completely amused by the present situation?”

1
0
0

@ptesarik @vbabka @piggo @corbet yes.

I will say something upstream, will put as softly and constructively as possible.

1
0
0

@ljs @vbabka @piggo @ptesarik @corbet "we should start with an AI policy document FIRST." - 100% on the money.

I'm also concerned that, whatever the intention, a CLAUDE.md file in the root of the kernel repo will be interpreted as endorsement of Claude specifically. And then every other LLM vendor will expect their own config file to be added to advertise their agent.

1
0
0

@ljs @pbarker @piggo @ptesarik @corbet two old men yelling at claudes

1
0
0

@vbabka @ljs @pbarker @piggo @ptesarik @corbet very eloquent branch-shaking djentlemen!!!

1
1
1

@lkundrak @vbabka @pbarker @piggo @ptesarik @corbet well it seems a civil conversation has yielded results!

Seems we will see an AI policy doc and have the maintainers summit to mediate this.

No unexpected merging should take place...

0
0
0