16:42:30 @rbrunner7:monero.social There has been a channel set up for automatic LLM reviews of certain pull requests - looks like it's found a small bug or two for your Polyseed PR 16:42:31 If you're interested, the room is #monero-slop:monero.social 17:42:06 It seems to be an empty channel on IRC. 17:42:23 Yes, it's not bridged to IRC (yet). 17:43:02 I don't think anyone has registered that properly and/or made the ask for that indeed 17:43:18 though... the bridge is probably going to just emit links to the long messages 17:43:30 Yeah, that's why :) 17:44:29 https://irc.gammaspectra.live/a5f642e6ab7b5ec7/image.png 17:55:19 ^ That's for Jeffro's corpus tests PR thing 18:43:34 I am really torn. On the one side, I want to deliver the best possible PR for Polyseed, with as few bugs as possible. On the other hand, I see a very slippery slope: Will it, in a very near future, be part of the "duties" of somebody making a PR, like I do now with Polyseed, to look at reviews done by AIs? If yes, how many dif [... too long, see https://mrelay.p2pool.observer/e/6ua6y6MLZTNud0py ] 18:44:41 It could turn out a bit like the breach of contract between reader and writer of texts if somebody just lets an LLM generate text. I look now at a review because I know a human did that, with a sense of duty, and it's my part of the contract to take it seriously. 18:44:50 How will that play out in the future, with AIs? 18:45:35 You make a good point, and yeah, forcing people to look at LLM reviews would be quite annoying 18:47:05 Someone made an essay recently about how AI is slowly turning open source projects into closing on themselves to avoid AI training and slop. "cash cows" can simply act as if they were not LLM and speak by themselves and you would look at the review without a second thought. Having to read slop, as in unreadable, uncomprehensiv [... too long, see https://mrelay.p2pool.observer/e/2-HHy6MLS1dhdkht ] 18:47:18 By asking DataHoarder: to not move spam from the room for spam into this room 18:47:22 Problem solved 18:47:23 If you don't want to, I can always look through it myself, and report any legit findings it might have come across 18:47:24 If someone does want it, they can join the room OR ask to be DMd it 18:48:11 @kayabanerve:matrix.org, i think rbrunner was asking about the near future of having to value or not llm reviews. not particularly this room potential annoyance. 18:49:08 It has found several legit issues with various open PRs which have been resolved since - which were overlooked my human reviewers previously 18:49:12 20:45:34 You make a good point, and yeah, forcing people to look at LLM reviews would be quite annoying 18:49:12 this is the same point I brought up on that channel, where even if the review is produced you are basically just shoving it into someone's face instead of bringing something up yourself 18:49:49 @jpk68:matrix.org: If you are ready to do that, that would be acceptable for me: Go through what the AI spat out, see what passes a first smell test and looks really interesting, post that in the PR as review comments. Because then you did work, so I will be glad to do work on my turn to check. 18:50:50 Please make sure to refer to the latest version if you do, I made changes Friday and today. 18:52:05 Yes but that doesn't change that in response to 18:52:05 What is our burden as we aware of potential? 18:52:05 I can give the snarky response,[... more lines follow, see https://mrelay.p2pool.observer/e/zPXZy6MLRDY2Z1Vm ] 18:52:20 I think it's beyond doubt by now that AIs do find interesting bugs and issues. But that does not mean it's already clear how duties, "best practices", workflows etc. will change in open source development with them. 18:53:45 Obviously, that doesn't scale to security issues, but I don't care to have a conversation on if we're morally obligated to use LLMs. I do care that I have the ability to be in this room and not see LLM slop. 18:53:48 I'm going to bitch at anyone who spams the room accordingly 18:53:54 Also, my bitchiness does have the actual comment "it should not be forced us, people should choose to engage", with no commentary on _when_ people should choose. 18:54:01 I'm refraining from that bullshit because I don't have the time and energy to deal with it 18:54:22 It's the same as, eg, coverity. It's about the felt probability to find something useful, the amount of work to go through it, whether you feel up to the task at that time, etc. 18:55:30 tsmt 18:56:40 Teenage Shaolin Mutant Turtles ? 18:57:10 this so much true 18:57:17 but nice try 18:58:32 In some way I am glad I am nearing the end of regular work and soon will only have to program what I really want myself. I think if you finish IT study today and have to find a job, in all this AI induced turmoil, man, that's hard. 18:59:20 @kayabanerve:matrix.org: Respectfully, I think we're well aware at this point of how much you dislike LLMs. Nobody here is trying to claim we should be allowed to shove slop in your face 18:59:58 I simply asked rbrunner if he thought it would be helpful, which especially applies for such a critical PR 19:01:55 People can use AI privately and then just submit the actual bugs they have verified. I don't think anyone can object to that. 19:02:31 tevador i don't object nor can i object 19:03:47 <@datahoarder:monero.social> I'd recommend that you bring specific issues reviewed manually and not AI output https://dontpastetheai.com/ you may need to do this manually somewhat. Most people won't want to engage with slop-related output 19:03:47 <@thomasbuilds:matrix.org> thing is this is hard to automate; so it ends up me reading AI work and verifying it (which takes time since I've not made the PR myself) and is a burden after some time 19:03:58 the TL;DR of the conversation on that channel :) 19:04:46 (I already run a LLM review anyway before adding something to the merge queue and add a comment if something looks worth pointing out) 19:30:32 @jpk68:matrix.org: I'm only bitching at people who literally do exactly that though. I feel like you're commenting I'm being overaggressive when afaict, I'm being fair. 19:30:40 I didn't say that you shouldn't offer as you did. I said DataHoarder: shouldn't post a link to a transcript (which auto embeds on my client) here. 19:31:37 ah. I think that was meant towards the bridge on the IRC side (when message is too long, see pictured example, it'd just send a link to IRC for the rest of the message) 19:32:16 19:42:06 It seems to be an empty channel on IRC. 19:32:16 even if the room was bridged it'd just be links in that room for whatever the slobot posts as seen on IRC side 19:32:54 it was an example of what would get embedded, not me trying to move the content here, sorry :) 19:34:35 jpk68 replied with a heart to selsta message which imply that he is a clanker support which imply that he should be EXECUTED. 19:48:43 @kayabanerve:matrix.org: I feel like you're commenting I'm being overaggressive when afaict, I'm being fair. 19:49:21 That's literally what I was trying to say. I didn't mean any offense, just letting you know :) 19:50:51 I called out an explicit person in response to a specific message, and it wasn't you. I hear if you feel the same though