02:09 <bitcoindev1337> hi i was blocked on github responding to https://github.com/bitcoin/bitcoin/issues/36761

02:09 <bitcoindev1337> i am not spam

10:09 <sedited> there is zero substance in that comment.

10:56 <stevenroose> yeah lol

10:56 <maflcko> Assuming willcl-ark is fine with that too: Would there be any objection of bringing the Ralph LLM review into a hidden/collapsed repo comment? The benefit would be to save one click and external tab for those that want to use it. For those that want to ignore the LLM-speak, the comment remains collapsed and can easily be ignored, just like the external link?

11:24 <sedited> I'd still prefer the link for now.

11:31 <sliv3r__> +1 on the link, I am using ralph a lot lately but I find it easier to find with a link as I know where it is, maybe looking for a comment in a long threat is not as good... unless we change the external repo link for a comment link :)

11:36 <maflcko> I was thinking it would be one of the earlier comments (like the Drahtbot one), so it would be easy to find, but still collapsed to be hidden by default. But no strong opinion. Making it an internal link is also possible, but seems fine to keep an external link for now. No rush to change it.

11:49 <willcl-ark> (of I am quite biased here but) I also feel like an inline comment would be nice at this point in time, seeing as ralph seems to now report few LLM hallucinations. This is already being trialled over in gui-qml for anyone who'd like to see how it looks, minus the collapsible comment part.

11:50 <willcl-ark> I think one reason for posting directly in the PRs is that it cuts out a reviewer having to post the (spammy?) "please check the ralph review" comment on PRs, and might also reduce reviewers re-posting findings ralph has already flagged.

11:51 <willcl-ark> sedited: might i ask for what reason you'd prefer not to see a direct comment on the PR? Whilst I get us not wanting tonnes of pure AI-written review comments, IMO having a single one per-PR wouldn't be _too_ bad...

11:53 <maflcko> My hope is also that having it collapsed, but inline would discourage all the other (duplicate) LLM re-hash reviews, but this is just a hope and could also backfire?

11:53 <sliv3r__> willcl-ark: that's a fair point tbh, I feel that a hidden comment + link to comment would be the best approach.

11:53 <sliv3r__> maflcko: a bunch of LLM posting suggested changes to fix ralph comments? lol

11:55 <maflcko> heh, it could also make it easier to detect and block the other re-hashing LLMs

12:08 <sedited> willcl-ark I think I just prefer a single comment for all our automated tools. It just seems easier to consume and less noisy to me. Plus might still be annoying to get to on a PR with many comments once github collapses them, even with a link to it.

12:12 <sedited> how about them in a collapsed section in the drahtbot comment?

12:12 <willcl-ark> That's fair, and there'd certainly be a "transition period" where with existing PRs, with many comments, ralphs comment would be at the bottom. whereas on new PRs it would be right under drahtbot (most likely). I'd still favour having them on the PR at this stage myself for the reasons outlined previously, but happy to go with rough consensus on the matter.

12:15 <willcl-ark> I think hooking up two two bots would be a bit janky. And would likely require something like ralph emitting webhooks for drahtbot so drahtbot knew when to update the collapsed section in it's own comment. I think I'd stick with the status quo vs that :)

12:27 <maflcko> DrahtBot should be back up today, so if there are any missing updates, lmk

12:40 <vasild> That is another story, but it would be nice if the PR author can somehow show to the others that he has read and acknowledged ralf's comments. Because now other reviewers have no idea if the author paid attention and dismissed, did not pay attention.

12:40 <vasild> _or_ did not pay attention

12:48 <maflcko> vasild: When the feedback is valid, I ususally say "thx ralph" or so. When it is ai slop, I ignore it as a test for other reviewers to see if the fall for the ai slop.

12:49 <vasild> :)