"""
Some of the AI reports we get are great, with no slop in sight and
finding really subtle and real bugs.

And others are very much not.

In general, I don't think that's all that different from bug reports
from actual humans ;^/

So I'd suggest just fixing the bugs that are noticed, and giving
credit appropriate to how good the report was.
"""
https://lore.kernel.org/all/CAHk-=wjcSt1gGnQZoyNvodky_6WEDxC=1+gQHywiOvOjw1+GUA@mail.gmail.com/

@irogers ugh it's his 'they're just tools like anything else' take in different clothing...

Suffice to say, I disagree :)

@ljs I was very impressed by this review coming on a patch from a maintainer: https://lore.kernel.org/linux-perf-users/20260408060017.9F21DC19424@smtp.kernel.org/T/#m014a667906a61b4e6441c251a2cc73c1a4de4897
Truly impressive breadth I've not seen from any other tooling.

@irogers yeah AI is good at review actually.

There can be a lot of noise too though, I've gone on about that, but in mm at least sashiko at about ~50% signal/noise in my expereince.

But having the tooling is a game changer, because AI does all the boring/painful stuff humans don't want to do, like figuring out if X then Y then Z might mean doing A with B is an issue.

In fact it can be 'superhuman' at this stuff, and esp. at things like diagnosing splats.

But it utterly sucks at creative things like generating code.

Plus it can hallucinate wildly so on both that and prior basis you need a human in the loop.

Given how LLMs work I don't believe that will ever change.

I use LLMs daily now, but I put them in their place, use them for what they're good at, see them as tools, and check what they output.

They're a force-multiplier when used right.

They're a slop generating pile of crap when used wrong :)

"""
But I think that lately quite a noticeable percentage of reports have
become very much valid. As you say, Sashiko tends to be quite good.
Even when it flags something unnecessarily and there's no actual
problem, there's likely to be a reason for the question.

And we've seen some truly stellar reports too. Things that are deep
and subtle, and the AI is 100% correct.

So anybody who thinks that all AI is slop is in denial about the current state.
"""
https://lore.kernel.org/all/CAHk-=wi_drr4Ls9KtXW1k8L2FUDF0YdnyjvKmPgLXHDFnnRWEg@mail.gmail.com/

@irogers There's loads of kernel code that's only understandable by 1 or 2 people anyway; so if the AI is spotting things that are actually difficult to understand those are worth improving.

@irogers Seems sashiko-bot is a bot that recently started working. Do you know more details about the bot including how I can use that? If so, could you please share those?

@sj So the bot does code reviews from Linux patches posted to the mailing list by scanning the mailing list using nntp. Sashiko has a web dashboard (https://sashiko.dev/) and recently we're mailing out code reviews on opted-in mailing lists (only to the mailing list and the patch author). The source code of sashiko is here: https://github.com/sashiko-dev/sashiko

Thank you @irogers ! Could you please further share how I can opt in, or check if I'm already opted in? I was thinking DAMON is already opted in [1] but I might missing something.

[1] https://lore.kernel.org/20260318150051.93173-1-sj@kernel.org

@sj Thank you for all your work too! I see DAMON as opted-in to emails here:
https://github.com/sashiko-dev/sashiko/blob/main/sashiko.dev/email_policy.toml
and in the scanned mailing lists here:
https://github.com/sashiko-dev/sashiko/blob/main/sashiko.dev/base/app/sashiko-k8s.yaml#L80