Conversation
monsieuricon

K. Ryabitsev

So, this is fun, but it really chews through your ChatGPT credits. :`)

I'm working on b4 "code review" mode (grab a series, apply it to your tree, review every patch, send all your acked-by's, reviewed-by's, and individual comments as a one lump batch at the end of your review). The reason I'm playing with this is to see if we can plug in some AI pre-analysis and discussion summaries before the reviewer starts their work.
1
4
9
vbabka

Vlastimil Babka

Reply to @monsieuricon
@monsieuricon what a ruined April fools opportunity to post that today?
1
1
4
monsieuricon

K. Ryabitsev

Reply to @vbabka
@vbabka Ha! :) I'm serious about the "summarize the reactions of follow-up reviewers," because this is what ChatGPT is actually okay at. But it may be too expensive to bother.
1
0
1
monsieuricon

K. Ryabitsev

Reply to @monsieuricon
@vbabka Here it is with just reaction summaries. Tell me this wouldn't be useful to see before you're looking at a patch series?
4
0
4
vbabka

Vlastimil Babka

Reply to @monsieuricon
@monsieuricon hmm maybe, depends on how much it would be hallucinating misleading stuff :)
2
0
1
monsieuricon

K. Ryabitsev

Reply to @vbabka
@vbabka One thing I'm not hallucinating is the "billed usage" chart, for sure. I spent $5 today just playing with a dozen or so series. Trimming message headers helps a bit, but this still chews up a ton of API tokens.
0
0
1
corbet

Jonathan Corbet

Reply to @monsieuricon
@monsieuricon @vbabka Sigh, I guess nobody will need @LWN anymore...
1
0
3
conor

Conor Dooley

Reply to @monsieuricon
@monsieuricon @vbabka ngl, reading that it makes me want a "b4 summarise v(N-1) -P <whatever is the dt-binding patch>!
0
0
1
monsieuricon

K. Ryabitsev

Reply to @corbet
@corbet @LWN @vbabka the value of LWN is not in summarizing random threads, but in summarizing *interesting* threads. I'll take your expertise in figuring out what those are over any automatic tool. :)

I see the value of ChatGPT here in summarizing dreadfully dull threads without worrying that it will burn out and ragequit.
1
0
4
ljrk@todon.eu

lj·rk

Reply to @monsieuricon

@monsieuricon @LWN @corbet @vbabka Also: Adding interesting context you might not know when just reading this thread, contextualising etc. -- but I don't wanna explain jons job to him, I just love the incredibly high quality of LWN. Thank you all :3

0
0
2
broonie@mastodon.social

broonie

Reply to @monsieuricon

@monsieuricon @vbabka I think one thing that would be really valuable in this area is “what are the people in this huge subthread arguing about?” and ideally “is it going anywhere?” but especially the second bit is hard and I’m not sure I’d trust AI. Or possibly spotting unusual review activity…

1
0
1
monsieuricon

K. Ryabitsev

Reply to @broonie@mastodon.social
@broonie @vbabka it's going to be an interesting trade-off between cost and convenience. Tokenizing huge threads costs quite a bit of money, but if it helps save time for maintainers, then it's a worthy trade-off. I will probably take a stab at incorporating this into the review workflow, but it will require some careful planning, including making sure that we don't lose the context of the tokenized discussion in case the developer wants to add to the prompt -- because we for sure don't want to re-tokenize the same huge thread and waste another bunch of money and cycles.
1
0
1
palmer

Palmer Dabbelt

Reply to @monsieuricon
@monsieuricon @broonie @vbabka IMO it's worth trying. One of my big issues is when someone keeps submitting patches, ignores core review feedback, and then just tires people out. Even just having a "this guy said your patch is totally broken and you ignored it 10 revisions ago to tire them out" blast when merging something would be awesome.

So it's not like I'd expect GPT to get everything, just to get ahead of the bad actors.
1
0
1
lkundrak@metalhead.club

HAMMER SMASHED FILESYSTEM 🇺🇦

Reply to @vbabka

@vbabka @monsieuricon then you'll just do the usual and review what the llm generated. review the review and that way it only adds a little extra work!

0
0
1
broonie@mastodon.social

broonie

Reply to @palmer

@palmer @monsieuricon @vbabka That’s the sort of stuff I have form letters for. There is one guy I got sufficiently frustrated with to simply stop sending him review comments but mostly it works.

1
0
1
palmer

Palmer Dabbelt

Reply to @broonie@mastodon.social
@broonie @monsieuricon @vbabka The problem on my end is that I don't always follow the previous versions of patch sets, particularly when a patch set touches multiple trees. So I try to go poke around the history to make sure nothing got missed, but that's a manual process and I tend to screw stuff like that up.
1
0
0
broonie@mastodon.social

broonie

Reply to @palmer

@palmer @monsieuricon @vbabka I simply don’t worry too much about what was going on with prior versions if I can’t immediately remember and review new versions afresh most of the time. I only go looking at changes if it’s something obvious that it feels like the submitter must be ignoring comments on, or if I otherwise remember something to check up on.

0
0
0
About social.kernel.org

Terms of service

  1. Please do not use this service in violation of the Linux Kernel Code of Conduct. Doing so will result in your account suspension with the referral of the matter to the CoC committee.

  2. "Repeating"/"boosting" someone else's status on this platform will be treated as endorsement and will fall under rule #1.

  3. You are encouraged to use this platform to promote your work on the Linux Kernel, but there is no restriction on permitted topics (with the exception of anything covered by #1 above).

  4. There is no requirement to post in English, but it should be considered the primary language of communication on this platform.

Privacy notice

The admins of this service have access to all posted statuses. They aren't looking, but if it's something they shouldn't know about, then you should not post it on this platform.

Please see the Linux Foundation Privacy Policy, which applies to this platform as well.

Getting your own account

If you would like an account on this instance, please check that the following applies to you:

  • You are listed in MAINTAINERS or CREDITS

  • OR: You have a kernel.org account or email address

  • OR: You have a long and established history of involvement with the Linux Kernel

If the above is true and you agree with the Terms of Service and Privacy Notice listed above, please use these instructions to request an account: