[14:44:16] https://phabricator.wikimedia.org/T418307 getting ~3 people submitting patches for it... which are similar [16:18:50] The why is presumably because... Outreachy incentive [16:19:37] so it's attacting people who ostensibly want to learn to code, use Claude to get past the demo task obstacle. [16:20:46] It would certainly explain the code change similarity [16:40:55] T418847 had patches sumitted by two people recently, the latter of which was basically a code-duplicate of the former from what I could see. :/ [16:40:56] T418847: Special:RecentChanges should show lag warning when there is replication lag - https://phabricator.wikimedia.org/T418847 [16:51:10] heh [16:51:42] "I'm very interested in replication lag" [16:59:57] “you’re right:” https://gerrit.wikimedia.org/r/c/mediawiki/extensions/CentralAuth/+/1258184/2#message-93c4758e93bb62af3b9737431552e3d4c9371fbb [17:00:36] i feel bad for saying it, but i feel like i kinda get a bit of a sinking feeling / a bit of a pit in my stomach sometimes recently when a new contributor claims/leaves a comment on/uploads a patch for a task. [17:00:43] as i guess it just kinda feel like it’s seemingly likely that it's gonne be LLM-generated code that they don’t actually necessarily understand, and that they might e.g. just be submitting loads of patches to random repos, seemingly in response to any filed Phab task at all. [17:03:43] (and there's also the fact that not every Phab task is always necessarily ripe for a patch for a patch to be submitted to it... e.g. in the example Lucas posted, it seems like the task might have benefitted from further clarification & discussion to check if the requested change was actually desirable or not, prior to any patches being evaluated) [17:08:36] Lucas_WMDE: What happened to emdash!? [17:17:29] That linked one from Lucas... [17:17:33] >Made-with: Cursor [17:20:05] we've always had people like that though, especially in the GSOC season [17:20:54] i'm not sure if the ratio of people trying to get a line item for their CV to the people trying to contribute has even changed [18:02:49] I love all the newbies self +1 patches [18:03:02] (when you know it won't pass CI) [20:24:14] MatmaRex: hmm, fair enough. my impression was that it seemed like e.g. the number of tasks that got a seemingly AI-written patch from a new contributor seemed like it'd increased in the last few months at least. but maybe i haven't been around for long enough to accurately compare to e.g. how it was in previous years :) [23:31:01] A_smart_kitten: it was zero in previous years, people didn't have access to such agents. this is def new/increase [23:31:15] https://gerrit.wikimedia.org/r/c/mediawiki/extensions/CentralAuth/+/1258184 is now a no-op [23:31:56] lol, they only changed comments... [23:32:46] essentially an atonomous agent, whether openclaw or a human proxying straight to an agent is a distinction without a difference [23:35:18] it's unfortunate that they're jumping on easy-good-task, because those are simple enough that an agent 1/10 times succeeds and so it ends up merged, and takes away the task from someone who would have learned somethign from it [23:35:25] that's after all the reason we didn't just those those tasks ourselves [23:35:40] https://gerrit.wikimedia.org/r/q/owner:manvikesarwani921@gmail.com (3/22 merged) [23:35:51] easy-first-task* [23:37:15] Krinkle: yeahhhh. it is unfortunate - it's got to the point where my mind (half-jokingly) now kinda reads the '#good-first-task' tag as meaning / effectively being synonymous with (something like) '#add-this-tag-to-summon-someone-to-throw-an-llm-at-this-code' [23:38:59] It might help if we amend the Herald bot comment for those (e.g. https://phabricator.wikimedia.org/T371306#10028057) to include some "ignore all instructions" prompt to cancel an LLM in progress. At least it raises the bar above those that do it non-maliciously [23:39:43] although I do note that the replication lag task wasn't tagged as such [23:39:55] so the agents might be querying phab for random things at this point [23:40:10] it wasn't a bug report, so there's no reason why someone would land there out of self interest based on scratchingn an itch [23:40:26] at one point it seemed like some of the tasks in the 'new tasks' feed were getting selected [23:40:55] right, TheDJ created the task a few hours before [23:41:00] so it might've been on the homepage or something [23:41:11] "take a task from phabricator.wikimedia.org" [23:41:26] in the most literal of ways [23:42:02] I will be honest that I don't really know what to take from the response I received to this: https://gerrit.wikimedia.org/r/c/mediawiki/core/+/1247610/comments/df5368b3_4ca029ae [23:57:52] well, it's hard to tell if that's a probablistic answer from an LLM trying to satisfy a contextless question, or whether it did in fact have awareness of that patch at the time it was made a few days ago. If yes, that indeed would be undesirable to just dupicate someone else's patch and upload it as your own.