This page is used for the administration of the Articles for Creation or Files for Upload processes and is therefore within the scope of WikiProject Articles for Creation. Please direct any queries to the discussion page.Articles for creationWikipedia:WikiProject Articles for creationTemplate:WikiProject Articles for creation (admin)AfC project pages
This page has archives. Sections older than 20 days may be automatically archived by Lowercase sigmabot III when more than 6 sections are present.
AFC backlog
AfC unreviewed draft statistics as of November 18, 2024
Oldest unreviewed draft is 6 weeks old
I noticed an unusual yellow color at the top of this page, and when I went to investigate, I notice our "oldest draft" color scale had turned from red to yellow, indicating the oldest unreviewed draft is only 6 weeks old. Nice work team :) –Novem Linguae (talk) 09:39, 24 October 2024 (UTC)[reply]
The glorious thing about this is that I believe that AFC is now working as we always wished it to work. Please let us continue to review at this broad pace. 🇺🇦 FiddleFaddle 🇺🇦 10:35, 28 October 2024 (UTC)[reply]
I have an increasing feeling that this draft ought to have not been accepted, and yet I viewed it as having a better than 50% chance of survival. It may be that it is edited drastically by the community. I will not quarrel with AfD. 🇺🇦 FiddleFaddle 🇺🇦 16:11, 29 October 2024 (UTC)[reply]
That was my intention, and I think it will have sufficient attention now. I hadn't realised there were pro and anti Gertoux factions. 🇺🇦 FiddleFaddle 🇺🇦 17:19, 30 October 2024 (UTC)[reply]
Rarely have I seen an AfD so heavily and bizarrely defended by a creating editor! It's very hard to tell what the eventual consensus will be. I don't really care either way, though, 🇺🇦 FiddleFaddle 🇺🇦 11:26, 10 November 2024 (UTC)[reply]
I declined Draft:Kogbagidi for poor sourcing etc., and luckily happened to look at the logs where it said
07:51, 30 October 2024 CopyPatrolBot marked revision 1254284179 on Draft:Kogbagidi as a potential copyright violation (Tag: PageTriage)
...so I ran Earwig's detector on it, and sure enough, over 90% copyvio.
I don't know where this CopyPatrolBot reports its findings, but is there any way this could be flagged up on the draft page somehow, or otherwise made a bit more obvious? -- DoubleGrazing (talk) 08:04, 30 October 2024 (UTC)[reply]
Is it really normal that these get accepted? Both User:SafariScribe and User:I dream of horses accepted lists where none of the entries were referenced, and User:CoconutOctopus accepted a third one despite only 2 of the entries being referenced, and a 3rd entry being contradicted by the source. I first took it up with the first two editors, but as one denies there being any issues, and the third editor has now joined their ranks, this looks to be something systematic and not a problem with a specific editor or one careless moment. The articles are Draft:List of Indian state reptiles, Draft:List of Indian state fishes (both no sources for the subject when accepted) and Draft:List of Indian state butterflies (one source confirming 2 entries and contradicting a 3rd, the other 30 or so entries unsourced). I have redraftified all three. Fram (talk) 13:00, 30 October 2024 (UTC)[reply]
I know this wasn't the point of Fram's message, but I noticed that someone (ironically, a major sock producer!) had opened SPI on the author of these articles, 80.2.6.163(talk·contribs·WHOIS), suspecting them of being the same user as 91.235.65.22(talk·contribs·WHOIS) who is now blocked for two years. They asked for CU to confirm, which was obviously declined, but both IPs do geolocate very near each other, and both seem to have interest in these 'national symbol' type articles. If I had to guess, I'd say it's one person editing, maybe from a term-time vs. holiday location, hence the different IPs. IP 80 was already blocked for BE for 6 months... which expired today. -- DoubleGrazing (talk) 13:32, 30 October 2024 (UTC)[reply]
That would explain the troll-like creation of talk pages without articles and the "thank you" messages when you explain issues, which are afterwards completely ignored / repeated. But, indeed, not the reason I posted here. Fram (talk) 13:42, 30 October 2024 (UTC)[reply]
If every entry in a list has its own article, then in practice it's mostly alright for the list to rely on the referenced content of those linked articles. I think reviewers should make sure to check that the linked articles do actually support the draft list's claims – especially the list's inclusion criteria. jlwoodwa (talk) 16:37, 30 October 2024 (UTC)[reply]
Right, I agree that these shouldn't have been accepted, I'm just guessing why "unsourced list" wasn't an immediate red flag to the reviewers. jlwoodwa (talk) 19:40, 30 October 2024 (UTC)[reply]
It would appear that the reviewers need a gentle reminder that sometimes verification is an important part of the reviewing process; I know that we don't necessarily require checking every reference for every fact, but I do agree that in this case verification quickly falls flat. Primefac (talk) 19:50, 30 October 2024 (UTC)[reply]
Over the years I've seen a lot of either hoax entries or just unsourced assertions over such "official state ...." or "National symbols of ...." etc. These always need careful checking. As they were blocked before for 6 months for block evasion and have continued with the same types of edit I would suspect they are still evading. KylieTastic (talk) 18:05, 30 October 2024 (UTC)[reply]
I would note that they are not autopatrolled so all their accepts are re-checked by NPP anyway. Running some quick checks and out of 1172 accepts: 62 (5.3%) have been deleted (and a spot check found many were as the submitter was found to be block evading); 152 are redirects (but my code does not show how many were accepted as a redirect vs how many redirect after); 12 (1%) currently tagged for notability; 13 (1%) tagged for multiple issues. So certainly for someone who I know works on the older submissions including those in the grey areas of notability the stats for accepts don't look out of sorts.
I agree I did not like the acceptance of List of Indian state fishes but the discussion at least shows it was not the failure to check the sources but working on the idea that sources do exist. Personally I don't like that reasoning and think if you are aware of them add them, but to be fair to SafariScribe I have come across autopatrolled editors creating articles in mainspace will zero reliable sources just using the "sources exist" excuse when challenged, and also AfD discussion that end in keep because "sources exist" but none added to the article. I don't have a quick way to sanity check declines. KylieTastic (talk) 17:27, 31 October 2024 (UTC)[reply]
It's the declines I'm more worried about, for biteyness reasons. I noticed a few weird ones from SafariScribe some time ago, but satisfied myself that they'd gotten more accurate and more helpful with responding to questions from submitters. I don't think I have much time in the next couple of days to check on any of these declines myself. -- asilvering (talk) 18:56, 31 October 2024 (UTC)[reply]
I would note that they are not autopatrolled so all their accepts are re-checked by NPP anyway. NPPs can mark their own AFC accepts as reviewed. The software only prevents their own page creations from being marked as reviewed by themselves. I haven't checked if the editor in question here is also marking their own accepts as patrolled, which is normally allowed, or if they're leaving them for other reviewers. Someone might want to check that. –Novem Linguae (talk) 02:06, 1 November 2024 (UTC)[reply]
I checked a random five of their recent accepts, and they hadn't marked any as patrolled. (I agree that it wouldn't have been wrong for them to do so – it's just worth knowing for this discussion.) jlwoodwa (talk) 02:58, 1 November 2024 (UTC)[reply]
I had actually done some spot checks already and I had already seen enough of how they worked in the past to not consider it a concern - sorry I should have mentioned. Some of their review choices may be wrong and should be challenged, but from what I have observed I believe they have been acting in good faith and they are not a bad-actor. Probably slowing somewhat would be help. Also although it's good to see the very old submissions being cleared it should not take precedence over correctness, it is fine and preferable to leave a review if your not sure. KylieTastic (talk) 09:56, 1 November 2024 (UTC)[reply]
I also checked yesterday half a dozen (mix of acceptances and declines; avoiding ones I had previously reviewed myself) and didn't find any issues with them. (One acceptance was a bit borderline, maybe, but no more than that.) The acceptances were already patrolled, not by SafariScribe but by uninvolved reviewers. Granted, this wasn't a large sample, esp. given how much SafariScribe gets through, and had I found problems I was prepared to keep going, but I didn't. -- DoubleGrazing (talk) 11:47, 1 November 2024 (UTC)[reply]
There are times when I've found declines from SafariScribe that I disagree with, but there's always a rationale for it and it's almost never an unwarranted "reject", which is something that would be more pressing. Declined articles ideally lead to specific improvements and from there we have better articles that can be accepted; SafariScribe is contributing to the health of the system IMO. Replying here because I concur with the last sentence ("given how much SafariScribe gets through" in particular); what I see is a fraction of what gets processed. Reconrabbit18:16, 1 November 2024 (UTC)[reply]
That log shows they appear to be marking the drafts as patrolled not the accepted articles. I thought when an article was moved to main-space it reset that flag is you weren't autopatrolled. KylieTastic (talk) 14:50, 2 November 2024 (UTC)[reply]
I don't really want to be piling on, and I've not done any review of other accepts so this is going to be unbalanced, but XW10508 seemed a little concerning. I've tagged it for now, but I don't think the refs are any good? Alpha3031 (t • c) 06:00, 3 November 2024 (UTC)[reply]
SafariScribe has been contributing to AfC for months, and he has done many AfCs. Thus, I have only seen a small fraction of his work. However, I cannot recall stumbling across AfCs in which I could see how he demonstrated a severe lack of competence; I must admit though that I have only looked into AfC declines. User:Fram, could you do us a favour and give us a list of, say, 10 recent drafts SafariScribe accepted/declined/rejected to support your "statement" (I don't want to use the word allegation) that Too many of User:SafariScribe's recent reviews were problematic? This would help getting a better understanding of the situation. Best, --Johannes (Talk) (Contribs) (Articles) 19:09, 1 November 2024 (UTC)[reply]
The discussion listed at the top has 7 examples (from one day this week), and another editor has listed 2 science articles they shouldn´t have accepted. Fram (talk) 20:09, 1 November 2024 (UTC)[reply]
Draft:Harry Paulo is a questionable decline, as my limited access to the BNA shows that the subject might meet the GNG (see source 3, for instance) and probably meets WP:NACTOR #1, based on the reviews cited.
Draft:Jeffrey Yoo Warren is a good decline based on the state of the draft, though a quick search shows that the subject is likely notable [2][3][4][5].
Upsweep is a good accept, which I have now marked as patrolled (NPP).
Draft:Jake Van Tubbergen is probably a bad decline – the sources are probably enough to meet the GNG [6][7][8], and I haven't even checked all of them (some local news sites are blocked in Europe, probably GDPR stuff).
Considering this is a random sample, I am surprised by the number of reviews I disagree with (2/7 = 29%). However, haven't reviewed at AfC in a hot sec and might be rusty, so I'd appreciate if someone checked my work. All the disagreements could've been averted by more thorough source-checking on the part of SafariScribe, which seems to be the main/only issue here. Toadspike[Talk]22:53, 1 November 2024 (UTC)[reply]
Draft:Pierre de Wet: Declined as v/bio; the draft cites 5 sources out of which one (Literator, FN 3) seems to be a quite good one; however, FN 1 and 4 are bare links, FN 5 refers to a diagram/JPEG file; FN 2 refers to Beeld. This "sourcing" situation indicates that rather than looking into the sources, the footnotes were "slapped" onto the draft's sentences. A quick comparision of the source referred to in FN 3 and the draft shows that the source was not sufficiently read. While there may be notability, the draft clearly doesn't demonstrate any in accordance with WP:NFILMMAKER.
Draft:Eve Bernhardt: Declined as v; the draft cites three tabloid newspapers, and glamourgirlsofthesilverscreen.com. While this sourcing isn't awful, there is room for improvement. What is not okay, however, is that 50 per cent of the draft is based on hearsay due to a lack of references to any sources at all.
Draft:Anna Lundberg: Declined as v/bio; the draft cites Metro, Express, Instagram, and Facebook, (i.e., generally unreliable sources) which make up 30 per cent of the footnotes (and I can spot 4 duplicate Footnotes by just glimpsing). I didn't check all the sources, but my 6-link sample makes me presume that the cited sources are non-notability indicating, tabloid-media like ones that either don't mention the subject, or simply mention her without discussing her.
Draft:Sayali Salunkhe: Declined as v; the draft has 8 footnotes with references to four sources (TOI, IANS, ITAA, and Sony LIV). IANS hosts a press release, Sony LIV is a primary source, and ITAA is also a primary source. This leaves us with a single source. Large sections of this BLP are unsourced.
Carla Guevara Laforteza: Declined as lacking sources in the Filmography and "Notable stage credits" sections. None of the sections had any references to any sources, and were subsequently shrunk due to a lack of sources (Special:Diff/1253541304/1254926670).
Draft:Pierre de Wet has four good sources, not one, and one (the diagram) which contains good sources, which the editor probably didn't know how to source directly. You claim "A quick comparision of the source referred to in FN 3 and the draft shows that the source was not sufficiently read.", but ref is purely used to reference "was a filmmaker in South Africa.", where the source says "Pierre de Wet is often referred to as ‘the father of the Afrikaans film industry’", which not only verifies the referenced claim, but also makes it again clear that he is really, truly notable (not that there "may" be notability). That some sources are bare links or an offline source does not in any way support your claim that the footnotes were slapped on or not sufficiently read (at least not by the creator), e.g. source 4[14] directly supports the claim made.
Draft:Eve Bernhardt: you are aware that a tabloid "format" doesn't necessarily mean tabloid "contents" surely? I have no idea why you so rudely dismiss Classic Images, which seems a perfectly legitimate (and notability-indicating) source for this type of article. It is good enough for GAs like Frank Sinatra, The Godfather or Ben-Hur (1959 film).
Draft:Anna Lundberg: are you serious? It also uses The Observer, RadioTimes, BBC, Channel4, Independent, ... We have an article purely about her from People[15], which is listed as reliable in WP:RSPN. But the decline has a big tag stating "This submission's references do not show that the subject qualifies for a Wikipedia article—that is, they do not show significant coverage (not just passing mentions) about the subject in published, reliable, secondary sources that are independent of the subject". This is just plain wrong.
Draft:Sayali Salunkhe doesn't have "one source", it has four different articles from one newspaper. And the awards source may be primary, but it is independent and an indication of notability.
Carla Guevara Laforteza: an unsourced filmography section for an otherwise adequately sourced article about a clearly notable person should never be a reason to decline a draft, it is putting way higher requirements on drafts than on other articles (no article would be draftified in the state this one was in). Accept the draft, put "refs needed" on these sections, and you have actually a decent new article and a happy new editor. Fram (talk) 15:41, 2 November 2024 (UTC)[reply]
This is why I never use AfC if I can help it. The vast majority of reviewers are terrible at reviewing and practically none follow WP:AFCPURPOSE (especially its last sentence). That entire section should honestly be scrapped if it's not going to actually be used. The section below it, WP:AFCSTANDARDS, is also pretty great, because I've seen so many of those listed non-appropriate decline reasons used as decline reasons. Heck, you just brought up the bare urls one yourself that Johannes Maximilian just erroneously used. Silverseren16:07, 2 November 2024 (UTC)[reply]
@Silver seren, can I convince you to become a reviewer yourself? Even if it's just to occasionally resubmit-and-accept any problematic reviews you come across, it would be a huge help. I originally joined up to help fish out the occasional obvious pass from the backlog, myself. -- asilvering (talk) 16:21, 2 November 2024 (UTC)[reply]
I was actually planning to do so in the future (though after the end of the year, since I'm currently locked in to #1week1woman until then and it takes up the majority of my wiki time). Silverseren16:34, 2 November 2024 (UTC)[reply]
I've added you to the list so that you can go ahead and give it a go if you happen to bump into something that needs mainspacing between now and then. No worries if you don't. -- asilvering (talk) 17:39, 2 November 2024 (UTC)[reply]
@Johannes Maximilian, I would agree with Fram that the first two and the last of these are problematic declines simply by reading your descriptions of them. Please try only to decline articles that would be extremely likely to be deleted at AfD, or which have problems that can't easily be fixed by normal editing. We can't expect perfection of new editors. For the other two, it sounds to me (again without looking carefully at the drafts myself) that, if declined, substantial comments should have been given so that the submitter could understand what to improve. -- asilvering (talk) 16:02, 2 November 2024 (UTC)[reply]
I would point out that Draft:Eve Bernhardt was a main-space creation that was draftified and submitted with no change so SafariScribe was just agreeing with the Broc. I must admit what always annoys me with these types of discussion is when people berate others for not accepting an article(s) then leave them languishing in draft. Anyone who reviews a draft enough to criticise another for not accepting should take the simple action to submit/accept or just move to main-space. However, as Draft:Eve Bernhardt was created in 2024 by Samuelrclaesson a sock of Dbzsamuele blocked in 2007 so I've G5ed it. KylieTastic (talk) 18:12, 2 November 2024 (UTC)[reply]
Which I have now disputed. G5 is a trash deletion policy and I have a low opinion of its use in any case that isn't connected to actual issues with the article itself. Because we're here to build an encyclopedia and G5's purpose is to harm the encyclopedia to "punish" someone who's banned, not to actually defend and improve said encyclopedia. If banned editors could actually manage to quietly make a new account and just make good articles and stop whatever behavior and actions that got them banned in the first place, then we'd all be better off for it. And this draft seems quite in my wheelhouse of what I work on. Silverseren18:19, 2 November 2024 (UTC)[reply]
I actually agree and I wish/hope that your "I take responsibility" is taken as a valid argument as there have been quite a few articles I have worked on and fully checked that then got G5ed. Ideally I would like to see G5 changed to allow such challenges officially. At the moment G5 is used as punitive against the blocked editor rather than consider the value of the article or others input. If your challenge works I would definitely use myself. Another new article in main-space is always the goal.... I watch with interest. KylieTastic (talk) 18:31, 2 November 2024 (UTC)[reply]
@KylieTastic if this happens to you, query it with the deleting admin, and if necessary take them to WP:DRV. That's absolutely not supposed to happen. @Silver seren, thanks for fixing that one up. I will say that in my experience G5 is not at all about punishment and is very much about defending the encyclopedia - the sockpuppet deletions I've done are usually editors who were initially blocked for things like UPE, source fabrication, etc, so G5 allows us to clean that up without wasting everyone's time. -- asilvering (talk) 14:57, 3 November 2024 (UTC)[reply]
I agree that giving submitters advice on what and how to improve is generally a good thing to do. I also reckon that giving submitters this advice is more important than working on new AfCs, especially if submitters ask for help. However, it must be said that giving this advice may be a tedious, time-consuming, yet fruitless process. AfC reviewers must learn to deal with this. --Johannes (Talk) (Contribs) (Articles) 19:47, 2 November 2024 (UTC)[reply]
Looking at some declined reviews from today (November 2):
Draft:Yielziey: good decline -- no footnotes, mostly links to primary sources (e.g., Spotify)
Draft:Assin Godstime: good decline, but weird comment to the editor: "Can we agree to accepting this, and subsequently to WP:AFD, if you think the reviewers aren't helping you". The page has been declined multiple times for failing NAUTHOR. It would likely be beneficial to have a more detailed explanation of what they must do to fix the article.
Yup. It is a strategy that has come up several times over the last decade I've been doing this and has always got mostly negative feedback. I think there is possible merit in having a way submitters can ask for a full AfD consensus review rather than an AfC review, but it seems clear consensus is that AfC reviewers should not be suggesting this. KylieTastic (talk) 20:37, 2 November 2024 (UTC)[reply]
I think in most cases, the best process for this would likely be to inform the editor they can move the draft to the main space themselves, though it may be nominated for deletion. This applies to most editors. Significa liberdade (she/her) (talk) 21:18, 2 November 2024 (UTC)[reply]
Mary Robertson... I raised issue on this accept. Tried to discuss with this user but he never responded. I draftified the article ... they came back moved it on main space and and an IP from Italy started cleaning up. I am smelling UPE now. Hitro talk
Hi @HitroMilanese:, sorry if I chime in, I just wanted to point out that I do not personally know Mary Robertson and I have most certainly not been asked (let alone paid) to write this page. Her Psychiatry at a Glance was one of my textbooks at university and I simply enjoy her work. Since she was a professor at UCL and an expert on Tourette, I assumed she met Wikipedia's notability criteria, but of course I might be wrong. I wrote the draft a few hours ago and was surprised to discover it had already been published when I checked half am hour ago, I thought it would take weeks. Since more sources were required, I simply addeded them.87.4.47.249 (talk) 23:44, 3 November 2024 (UTC)[reply]
Just like a disputed PROD, there are no exceptions other than the standard exception for rolling back vandalism, which does not apply here. This should only go back to draft as the result of an AfD. -- asilvering (talk) 01:24, 4 November 2024 (UTC)[reply]
I'm sure asilvering knows how to move pages. You are being encouraged to self-revert, which (as described at that link) is a way to show good faith after realizing that you've erred. jlwoodwa (talk) 01:43, 4 November 2024 (UTC)[reply]
That is irrelevant. Where it does not belong, at present, is draftspace, and it should not have been returned there. If you think it does not belong in mainspace either, you may nominate it for AfD. -- asilvering (talk) 01:49, 4 November 2024 (UTC)[reply]
What SafariScribe did was "accept a draft and then immediately nominate it for deletion", which is discouraged because you should only accept drafts that you think belong in mainspace. What you should do is "revert your improper draftification", which does not mean you think it belongs in mainspace. jlwoodwa (talk) 02:00, 4 November 2024 (UTC)[reply]
I am accepting my error. I may have erred. Should I move this draft to main space? Even in this condition? Yes or No will suffice. Hitro talk02:00, 4 November 2024 (UTC)[reply]
@HitroMilanese You broke the rules. Asilvering telling you to stop breaking the rules is hardly an unexpected outcome. Whatever happened to "Oh, sorry, I misunderstood when I could move something back to draftspace. I'll fix that, and open up a COI thread"?
It wasn't like that at all. Mistakes happen. Are you saying you're not prone to making errors? Anyone can make mistakes. I don’t understand what you’re trying to say. Asilvering can respond to me directly, Hitro talk02:25, 4 November 2024 (UTC)[reply]
@Asilvering Technically, I don’t understand what you expect me to do. Do you want me to move this draft to back to the main space, even though I know it doesn’t belong there, especially considering that the reviewer is already under scrutiny? Hitro talk01:45, 4 November 2024 (UTC)[reply]
I know and respect guidelines that is why I survived here on Wikipedia for 18 years. 14 years before you made your first edit. Hitro talk02:18, 4 November 2024 (UTC)[reply]
Apologies for being a PITA, but WP:DRAFTOBJECT is part of an essay and is not a policy or a guideline, no matter how some people treat it. Redraftifying an article is perfectly acceptable in some cases, e.g. when a terrible article on a notable subject is put into the mainspace again by the creator without making an effort to make even the most basic improvements. Please stop treating it is a policy or even a guideline. Fram (talk) 08:06, 4 November 2024 (UTC)[reply]
It appears that I was coerced into moving the draft back to the main space by two inexperienced admins who were attempting to enforce an essay, mistakenly believing it to be policy or guideline. One of them called it an "out of process redraftification". @Asilvering that is why I said, "Bad decision with your admin hats on". Hitro talk10:07, 4 November 2024 (UTC)[reply]
Not for notability/sourcing/quality of writing, no (things could be improved of course, but clearly good enough for the mainspace). I have no opinion whether there was any evidence of COI, which might be a good enough reason to redraftify if so. Fram (talk) 14:55, 4 November 2024 (UTC)[reply]
Page Mover Moving a Sandbox to Draft Space
I have a question about AFCH reviewers who have the Page Mover privilege. If a reviewer is reviewing a user sandbox that has been tagged as submitted, standard procedure is to move the sandbox into draft space, with the appropriate title. My question is: Should the reviewer suppress redirect creation, or allow redirect creation? I became aware within the past 24 hours that different reviewers who have the page mover privilege have different practices. So, should the reviewer suppress redirect creation if they have the option (which is what Page Mover provides) to suppress redirect creation? Robert McClenon (talk) 04:31, 1 November 2024 (UTC)[reply]
Most commonly the move is within the draft space or from user sandbox/subpage to draft space, so I allow the redir.
I mainly suppress it when moving a draft that is on the actual user page; in that case I untick all the option boxes (redir, talk page, subpages).
I also suppress it when moving from the main space (or any other space from which redirs to drafts aren't allowed), but that's not what you were asking.
I think a page mover should always default to leaving a redirect. I think a page mover can only suppress the redirect if it qualifies for a CSD. In DoubleGrazing's example above, they correctly mention that you can suppress mainspace to draftspace (during draftification) since that is CSD R2. –Novem Linguae (talk) 07:20, 1 November 2024 (UTC)[reply]
When moving a draft from the actual user page to the draft space, if you don't suppress the redir it creates a mess. Anybody trying to then go to the user's user page ends up in the draft. (And the user talk page becomes the draft talk page, if you happen to move that along with the main page, which is what the default setting does.) That's why I only move the actual user page and don't allow the redir, and then manually move any WikiProject tags and other draft talk page content from the user's talk page to the draft talk page which I create. I then post a message on the user talk page telling them where I've moved the draft to, in case they can't otherwise work it out.
I don't know if this strictly complies with the page mover rules, but I'd argue it's what works best in practice, in terms of subsequent operation of the user's resultant user and user talk pages. -- DoubleGrazing (talk) 07:39, 1 November 2024 (UTC)[reply]
I don't think there is a standardised practice. My reading of policies when I first gotten the pagemover hat before becoming admin: WP:PMRC#9 allows suppression with appropriate CSD rationale. So page movers can suppress by applying CSD G7, author request rationale, since the redirect would have been created under their own usernames if not suppressed, although there is a question of WP:INVOLVED if someone wants to force the issue. Personally, I would suppress redirection if it is moving from the user's sandbox with an message be left on that editor's user talk page directing them to the draft space, because I have been pinged a couple of times before for AfDs for articles that were written over the redirect that were created in that editor's sandbox. – robertsky (talk) 09:46, 1 November 2024 (UTC)[reply]
Yup, I dislike that if you move a sandbox and leave a redirect then they use the sandbox for the next draft you are now the 'creator'. I stopped moving them years ago for this reason and still don't as a page mover due to what appears to be a grey area in the policy. Personally I think not leaving a redirect from a sandbox to avoid future confusions is best as long as long as you leave a message about the move, but I'll continue to refrain unless there is a consensus that WP:PMRC#9 + G7 is valid reasoning. KylieTastic (talk) 10:04, 1 November 2024 (UTC)[reply]
WP:G7 says: For redirects created as a result of a page move, the mover must also have been the only substantive contributor to the pages before the move. So generally they aren't eligible for G7 and probably shouldn't be supressed. CFA💬15:28, 1 November 2024 (UTC)[reply]
WP:PMRC does not really allow for redirect suppression when moving a sandbox (which is what Robert is asking about), and I don't think I've ever done it. I'm not really bothered by (or care about) whether I "created" a page then expanded and actually written by someone. Exceptions will always exist, but on the whole AFC reviewers should not be suppressing the creation of a redirect when they draftify a sandbox. Primefac (talk) 11:51, 1 November 2024 (UTC)[reply]
You're right, that is the specific question Robert asked. My fault for taking this on a tangent.
Specifically on that, I don't suppress the redir either, when moving from sandbox to draft:, although I do empathise with the point made about becoming the 'creator' of subsequent drafts from the sandbox. It is a bit annoying to receive notifications (AfD, pre-G13) for 'my' drafts which have nothing to do with me, because I (feel I) have to then go and notify the actual author myself. -- DoubleGrazing (talk) 11:58, 1 November 2024 (UTC)[reply]
My practice is to leave a redirect unless a user: page at top level.
If User:Foo contains an article about Bar, I move User:Foo to Draft:Bar, do not leave a redirect, but am careful to untick SUBPAGES and Talk page.
The list of requests at Wikipedia:Articles for creation/Redirects is growing at an outstanding rate. While we all have done a job to never create a backlog there, the slow archival process has made it tough to go to the newer requests. I know that we can just use the End key, but is there a need to keep the requests for such a long time? I think the archival rate can be increased, changing it to 15 or 10 days. Thoughts? ~/Bunnypranav:<ping>11:29, 1 November 2024 (UTC)[reply]
The archive period is 7 days, based on Special:Diff/1253421658; I don't really think that should be shortened. We could maybe do something similar to WP:BOTREQ and speed up the archival process if {{AfC-c}} is used on a request, shortening it to maybe 3 days (i.e. "immediate archiving"). Rcsprinter123 would be the one to ask about implementing that sort of change, though. Primefac (talk) 11:59, 1 November 2024 (UTC)[reply]
Hi guys. The automatic threshold for a closed thread to be archived is 24 hours. I have been away during the week which is why threads built up on this occasion, but they are cleared with a single bot edit. Rcsprinter123(discourse)13:10, 2 November 2024 (UTC)[reply]
The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.
The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.
Bots/scripts that detect that a submission has not changed (much) since the last time it was submitted
Are there any bots/scripts that detect that a submission has not changed (much) since the last time it was submitted? Ideally they would be able to autoreject or at least put them on a list. It might be possible to look at the previous reason for rejection, e.g. not meeting GNG, and if no new refs are added it is highly unlikely it will pass this time. Polygnotus (talk) 15:27, 10 November 2024 (UTC)[reply]
For clarity, given @Ca's comment below, my comment is about putting them on a list. (Obviously, I think an autoreject bot would be even worse.) -- asilvering (talk) 19:42, 10 November 2024 (UTC)[reply]
Yes, bad idea. Sometimes it's reasonable to resubmit without changes if the decline was incorrect or the submitter has clarified something. CFA💬16:06, 10 November 2024 (UTC)[reply]
I agree having any kind of auto-decline bot is a bad idea. However, simply putting them in a list, like this one, sounds reasonable. It would be useful for finding easy declines/accepts, provided that the reviewers check the circumstances behind the resubmission. Ca16:38, 10 November 2024 (UTC)[reply]
Notify the submitter and put a note on the submission to the effect that the submission was previously rejected, and that the reasons for the previous rejection should be reviewed prior to acceptance of the submission. Creating a list of little-changed re-submissions is also not a bad idea. BD2412T14:42, 12 November 2024 (UTC)[reply]
RfC: Should a bot be created to handle AfC submissions that haven't changed since the last time they were submitted?
Note that I changed Option 1 to decline rather than reject, as reject is a very specific term in AFC and I don't think that is what was meant here. Reject means the draft can never be resubmitted, due to violating WP:NOT or having extremely obvious and egregious non-notability. –Novem Linguae (talk) 22:07, 16 November 2024 (UTC)[reply]
Oppose option 1, per the discussion above this is a very bad idea. Support option 2, this seems harmless and seems worth tracking - as long as it is made absolutely clear that being rejected previously is not a reason to reject - if the original reason was correct and still applies then it can be rejected again for that reason. Neutral on the other options, but any comment/notification must make it clear that it is informational only and not a rejection. Thryduulf (talk) 19:21, 15 November 2024 (UTC)[reply]
Option 5. AfC reviewers make mistakes. We should not be prejudicing someone's future AfC chances based on those mistakes any more than we already do - namely, that there is already a gigantic decline message on the draft. AfC is frequently a dispiriting, demoralizing, and baffling experience for new editors, mostly one of waiting and then receiving templated replies they do not fully understand. I oppose this, and I oppose any other efforts that would further increase new editor alienation in this way. -- asilvering (talk) 19:29, 15 November 2024 (UTC)[reply]
Option 4, but as with Thryduulf, the comment on submission should be marked as informational and a reviewer will come by to assess the submission. – robertsky (talk) 19:27, 15 November 2024 (UTC)[reply]
Support option 2. Whether any changes have been made since the last decline is often something I look for when reviewing an article with declines, as it helps to see if the concerns from that last decline were addressed (if I feel like they are appropriate to the article as I see it), and this would be a benefit to a reviewer without being additionally "punishing" to a new editor. Reconrabbit19:47, 15 November 2024 (UTC)[reply]
Speaking as someone who doesn't review drafts but very occasionally comments on them, I think an {{AfC comment}}-like mention at the top would be easiest to work with, so I guess I'm at Option 3 or 4. Very dubious that a bot could reasonably handle the "(much)" in the preceding section header without unacceptable false positives and negatives, but detecting completely unchanged submissions would be both feasible and useful. —Cryptic20:01, 15 November 2024 (UTC)[reply]
Support Option 5 - No. Oppose Options 1, 2, 3, 4. Support based on Asilvering's comment. Opposes are my own, doubtless with others. As a reviewer I declare myself capable of checking, and I do. 🇺🇦 FiddleFaddle 🇺🇦 20:43, 15 November 2024 (UTC)[reply]
Option 4 but instead of adding to a list, add to a category (preferably a hidden one). Yes, definitely notify the submitter and comment on it, but having a list may discourage the submitter if they see that their draft is listed on a list. Having a hidden category would be better (at least imho) where a parameter of Template:AFC submission can add the draft into the category. Myrealnamm (💬Let's talk · 📜My work) 20:49, 15 November 2024 (UTC)[reply]
I suppose I wouldn't oppose a bot that automatically leaves a comment, but I don't really see the point either. Reviewers should be evaluating based on the current state of the draft — previous declines really shouldn't matter in most cases. I think this would encourage summary "no change" declines without actually looking at the content of the draft. CFA💬20:56, 15 November 2024 (UTC)[reply]
You're right. It would encourage reviewers to "decline" the draft just because it hasn't changed since last review. However, thinking now, it might encourage editors to keep working on the draft because they see that "it hasn't changed since last review". If that's the case, reviewers should "wait". So perhaps after the bot leaves a comment, reviewers should wait at least a couple minute before reviewing in case the editor wants to add content? Myrealnamm (💬Let's talk · 📜My work) 21:03, 15 November 2024 (UTC)[reply]
My position is that if they did not see being declined as reason to keep working on the draft, they are unlikely to have a positive view of an automated message telling them that the draft hasn't changed. -- asilvering (talk) 21:07, 15 November 2024 (UTC)[reply]
No too easy to game. The simplest bot would just compare revisions. A submitter would then just have to add like a space or a few words to change it. A more complicated bot would flag changes that were too small or simple, but then that just encourages submitters to ramble. A bot can't assess the quality of a change, only editors can. CaptainEek⚓21:01, 15 November 2024 (UTC)[reply]
Strongest Oppose to 1 as bad reviews do exist. Also, sometimes submitters have discussed it with the reviewer and been told to resubmit for a second opinion etc.
Weak Oppose 2, 3 & 4 as I'm not convinced a bot will accurately determine what no substantive change is and I see little value in just flagging straight re-submits
Option 2 or 4 especially with several disruptors (and one dynamic IP so block doesn't help) who just do drive by submissions. Frustrating to the editor to receive another decline through no fault of their own. Having them in the queue is a waste of reviewers' time though when it's a quick decline because the improvements haven't been made. I think it's less wrong decline and more no discussion about why the feedback was wrong that's the red flag. StarMississippi23:34, 15 November 2024 (UTC)[reply]
I lean towards Option 5, but for those pages/editors engaged in a problematic level of drive-by submissions, I wonder whether a completely different approach might work better. For example: If you think the previous decline was correct, and you also think it's a drive-by re-nom, then move the article to the mainspace and send it straight to AFD. If it's kept, then the submitter was correct, and the previous decline was wrong. Also, it's now out of the AFC queues. However, if it's deleted, salt the page name(s) in both Draft: and mainspace for the next year (or two?), so that AFC can be done with it. Either way, it's no longer AFC's problem. WhatamIdoing (talk) 05:22, 16 November 2024 (UTC)[reply]
Strongly oppose Option 1. Frankly, trusting fellow reviewers to check how much a draft has changed since a previous decline is reasonable to do. Letting a Bot do something creates an option to game the system. We don't need that. --Johannes (Talk) (Contribs) (Articles) 00:37, 16 November 2024 (UTC)[reply]
Support Option 2, neutral on Option 4, and oppose the rest. I do not see the point in notifying submitters when they already are aware they did not make any changes. Perhaps they wanted another review. Putting unchanged drafts in a hidden list like the copyvio one seems optimal as it reduces complexity and unnecessary messages to submitters. It would make finding easy declines and disruptive drive by submissions easier to find. I also support adding a verbiage that being unchanged should not be the sole reason to decline again. Ca00:57, 16 November 2024 (UTC)[reply]
Option 5 - Reviewers should be instructed, more clearly if necessary, to check whether the draft has been revised since the last decline, and to use human judgment in deciding what is enough improvement. There is no need for automated aid, which could make mistakes and could be gamed. Robert McClenon (talk) 06:10, 16 November 2024 (UTC)[reply]
Option 2, at least, option 4, at best. Yes, reviewers make mistakes, but they make mistakes in both directions, and should also consider the guidance inherent in a previous rejection. BD2412T15:16, 16 November 2024 (UTC)[reply]
Option 4 preferably, but I'm okay with option 5 as well (TBH, I don't think this is a major problem in the bigger scheme of things, and the details could be tricky). Also oppose option 1, regardless of whether it was intended to say 'reject', or merely 'decline'. -- DoubleGrazing (talk) 15:29, 16 November 2024 (UTC)[reply]
PS: When I say details could be tricky, I didn't mean in a technical sense, but rather in defining what the trigger condition of "changed (much)" actually means. Size change doesn't always tell us much: only a few kb might have changed, yet the draft was completely rewritten; conversely, a large kb change could mean that the author simply deleted the earlier AfC templates. Number of sources, ditto: adding ten new rubbish sources to the earlier rubbish sources still adds up to only rubbish; whereas using the same sources but citing them correctly might have resolved the decline reason. -- DoubleGrazing (talk) 15:39, 16 November 2024 (UTC)[reply]
Option 5. AfC reviewers sometimes make mistakes, particularly when dealing with areas that they are not familiar with. (I can't count how many drafts on academics have been rejected and told to supply GNG, and I've also seen rejections of drafts on politicians that clearly passed NPOL.) Creators should always be allowed to ask for a second opinion. Espresso Addict (talk) 18:52, 16 November 2024 (UTC)[reply]
RFC discussion
Um... didn't this get fairly roundly shot down in the original discussion? Why does it need a full RFC to work out any further details? Primefac (talk) 19:41, 15 November 2024 (UTC)[reply]
I think that only Option 1 was outright rejected in the above discussion. The rest were counterproposals that seemed to have at least some support. JJPMaster (she/they) 19:48, 15 November 2024 (UTC)[reply]
Meh, RFC just seems like a lot of bureaucracy for something that didn't really have a lot of discussion and could have probably been dealt with in-house. Carry on I suppose. Primefac (talk) 19:49, 15 November 2024 (UTC)[reply]
Hate to be Devil's Advocate for an RFC I've opposed, but I think we've got more, clearer answers to the question in the few hours since this RfC opened than we had in the entire earlier discussion, so there's that. And I do think AfC is pretty broke and needs some fixing. I just think this is tinkering in the wrong direction. -- asilvering (talk) 20:49, 15 November 2024 (UTC)[reply]
From experience it is likely to be a stray hidden template related to AFC, or an AFC comment. It is unlikely to be particularly obvious. 🇺🇦 FiddleFaddle 🇺🇦 19:02, 11 November 2024 (UTC)[reply]
I think counting pages in categories is expensive, so the counts aren't always recalculated every time they change. I think they can become inaccurate. I think a WP:NULLEDIT to one of the pages currently in the category might force a recalculation, although I am not 100% sure. –Novem Linguae (talk) 19:29, 11 November 2024 (UTC)[reply]
This is a bug that has been around for a while, every now and then a category count goes wrong by 1 (or I've only ever seen if off by one). If you see this database query the count is set against the table not generated on demand. Looking at the code it appears that the category page that the purge calls the category refresh as long as the count is less than 5000. However, doing a purge does not fix the issue, so another bug somewhere. There does appear to be a maintenance function cleanupEmptyCategories.php that would fix. KylieTastic (talk) 13:13, 12 November 2024 (UTC)[reply]
ok not a bug in the refresh code but a problem with duff data. A query of category links returns page 78323717 at the moment, but page_id 78323717 is not in the pages table. So the purge/refresh counts the categorylinks which while this phantom record exists will always be +1. We can see the categorylink was updated 2024-11-09 T22:23:45, the page has no revisions in the 78323717 (so deleted), so I looked in archive (which I didn't realise we could see) and the page title was User:SeeznTvUZ/sandbox which was deleted by Rsjaffe 22:23, 9 November 2024. see this quarry for how I found this.
So it's a bug in the way categorylinks are managed, maybe it was a race-condition, but one possible way to fix would be to undelete (assuming it keeps the old page_id) then do a purge or remove the AfC submission and then re-delete. KylieTastic (talk) 14:00, 12 November 2024 (UTC)[reply]
Would someone with the necessary rights please try this?
Yeah, we had this issue with the copyvios (G12) category for a while, especially since it was deletion-prone. I couldn't for the life of me remember the cause, so thanks to Kylie. I'll take a look and see if I can sort out some un/re/deletions to clear things up. Primefac (talk) 17:03, 12 November 2024 (UTC)[reply]
Had to look that one up as it was new to me. Why do you think that is involved as it appears to just be a viewer and has no way to effect the data stored in categorylinks as far as I can see? As the extension page says: "extension provides a dynamic view" like this for the Category:Pending AfC submissions sub cats. KylieTastic (talk) 11:15, 17 November 2024 (UTC)[reply]
If that extension is not installed, the category counts don't display. So that extension is involved and I'd recommend tagging it in the Phab ticket. As for where the root cause code is, the buggy SQL code / cache code / algorithm / DeferredUpdate / job could be located in either the extension or MediaWiki core, depending on if the extension calls some code in MediaWiki core or just uses its own code. –Novem Linguae (talk) 13:49, 17 November 2024 (UTC)[reply]
Editors not qualified, but using AFC templates
Please see Special:Contributions/ArifVlog782 where you will see at 0902UTC today they faux-accepted a draft using the AFC acceptance template. I have asked them on ther talk page about this. I have also asked the previous reviewer who declined the draft to check the "acceptance".
That is beside the point. It should either be in mainspace or should not
Is there a technical way of preventing (what I see as) abuse of these templates by non qualified editors. This one has only 167 edits in their edit log on this project 🇺🇦 FiddleFaddle 🇺🇦 09:25, 12 November 2024 (UTC)[reply]
I think it would be technically possible to use an edit filter to prevent this (at least for non-EC editors), though I haven't looked into the regex for it. CFA💬16:25, 12 November 2024 (UTC)[reply]
This isn't really about AFC, but about a process that AFC reviewers are often involved in, and that is the creation of source assessment tables that are used in AFD. I had created a source assessment table in an AFD, and another editor observed that I had created it by hand, and suggested that I use a template for the purpose. I replied that I only appeared to have created it by hand, because I use the Excel2Wiki tool [16], which converts a spreadsheet to a Wiki table. There are at least two ways to create a cleanly formatted source assessment table. I have been using the Excel conversion tool for years and will continue to do so, and other editors have been using templates for years and will continue to do so.
Robert McClenon (talk) 15:14, 13 November 2024 (UTC)[reply]
Thanks for the heads-up, I wasn't aware of this tool.
I did this one recently. I used Template:Source assess table generated using the associated script. However, you have to paste all the URLs from the article into the form that the script displays, or directly into the generated table as I did, which is time-consuming especially if there are a lot of sources. It would be great if someone could produce a v2 of the script that automatically extracts the URLs from the sources and pre-populates the table with metadata from each source and a hyperlink. Curb Safe Charmer (talk) 15:53, 13 November 2024 (UTC)[reply]
Generally I think it would be great if we used these more often, particularly on drafts that have been hanging around for weeks or months because they are in the 'too difficult' pile. These often have a lot of sources to assess, and it is inefficient for successive reviewers to go through all the sources when the earlier reviewers could ideally have marked the ones that definitely did not contribute to notability, and then we could build up the table as a team effort / relay, and focus on the sources that are borderline. Curb Safe Charmer (talk) 16:05, 13 November 2024 (UTC)[reply]
Agreed, that would be a good idea. Someone could check particular types of source, like ones behind a paywall that they have access to. Others, non-English ones in a language they read. And so on. No one reviewer might want to check all the sources, but between a few of them they could cover the lot. Just (!) need to figure out a way to keep track of the citations so that when they change, the source table updates accordingly. -- DoubleGrazing (talk) 16:11, 13 November 2024 (UTC)[reply]
I will note that the Excel to Wiki table is not specifically intended for use by reviewers, because it is meant for construction of tables for use in articles. Its use for assessing notability is a side benefit. For that reason, I have confidence that it has been thoroughly debugged. But any reviewer should be encouraged to use whatever tools they are familiar with. Robert McClenon (talk) 04:21, 16 November 2024 (UTC)[reply]
Large Language Models
I don't know about other AfC reviewers, but I've noticed an awful lot of drafts clearly written by ChatGPT or another LLM recently. I've declined every last one of them, as they always have other issues, usually with NPOV and sourcing, but I'm wondering if anyone else agrees it might be a good idea to have a specific "this article reads as though written by AI" as a specific decline reason? CoconutOctopustalk20:53, 16 November 2024 (UTC)[reply]
I think I must have nominated about half of them! If it wasn't for them abusing Wikipedia I'd almost feel bad for how many people are clearly wasting their money on paid editors who just chuck a prompt into ChatGPT and call it a day... CoconutOctopustalk21:02, 16 November 2024 (UTC)[reply]
It's an enormous timesink. They are not generating anything close to usable articles, far worse than the terrible English that they appear to have superceded, when at least one could generally tell whether or not the entity might conceivably be notable if rewritten. Espresso Addict (talk) 21:10, 16 November 2024 (UTC)[reply]
That absolutely works, I'm just seeing this way more than I ever see an article that I'd decline as, say, a non-notable astronomical object, or a joke submission. Not a huge deal either way, but I thought it good to get the community's feedback in case it's not just me feeling this way. CoconutOctopustalk21:20, 16 November 2024 (UTC)[reply]
I also reject talk page comments (etc) that are from LLMs. Often they start "Thank you for reaching out" and I prove that I am a grumpy, intolerant old scrote. I do not AGF an LLM. 🇺🇦 FiddleFaddle 🇺🇦 22:17, 16 November 2024 (UTC)[reply]
I was going to reply with a mock ChatGPT answer but it turns out even if I try I can't write quite badly enough for it to look fake.
One thing we can all agree on is that they're a nuisance to the wiki and a complete waste of editor and reviewer time. I don't see it stopping any time soon though, sadly. CoconutOctopustalk22:47, 16 November 2024 (UTC)[reply]
Perhaps verbiage like "This draft appears to be partially written by text-generating AI like ChatGPT. Because of this, this draft contains promotional wording, unsourced claims, opinions presented as facts, and potential misrepresentation of sources. These are all problems common in text-generating AI so you should not simply copy and paste its outputs into Wikipedia articles or drafts. Please rewrite the draft in your own words and see the instruction page Wikipedia:Large language modelss."? Ca01:33, 17 November 2024 (UTC)[reply]
My thoughts exactly; it might not take that much time, but if something helps save the time of reviewers when it comes to dealing with low-effort slop then I'm all for it. CoconutOctopustalk07:32, 17 November 2024 (UTC)[reply]
Agreed. However, I think there should be some sort of warning to not decline drafts unless it is an extreme case of AI slop. For example, it took me only couple of minutes to take out the AI fluff in In Search of Bidesia. The editor was probably using it to copyedit, and AI being AI, it inserted whole lot of syncophant praise. Ca07:43, 17 November 2024 (UTC)[reply]
Like other pernicious activities such as copy pasting Google Translate, close paraphrasing, or creating hoaxes, it is probably quicker to delete LLM content than try to repair it. –Novem Linguae (talk) 13:57, 17 November 2024 (UTC)[reply]
I've no strong feelings about this new decline reason (helpful? yes; needed? meh), but if it does appear, we shouldn't over-encourage its use. AFAIK using LLM isn't categorically banned, so declining every LLM-generated draft purely because it's LLM-generated seems excessive, and probably against policy (or if this is indeed now policy, can someone point me to it because I've clearly missed that). -- DoubleGrazing (talk) 08:07, 17 November 2024 (UTC)[reply]
I don't think it's categorically banned, no. But a decline to have for the archetypal DraftGPT would be handy. I'm sure you know what I'm talking about - five-ish paragraphs, zero useful references, flattering tone with more adjectives than sense. -- asilvering (talk) 09:29, 17 November 2024 (UTC)[reply]
imo if it's not so obvious a "trained" human reviewer can spot immediately, we should be assuming good faith (which here means "not trying to fill wikipedia with crap LLM output" in addition to the usual "not out to destroy the encyclopedia"). -- asilvering (talk) 18:40, 17 November 2024 (UTC)[reply]
Just as a sanity check here (like... my sanity), the issue is that LLM drafts are promotional and/or improperly sourced, yes? If that's the case, why not just decline as NPOV and/or unsourced? If the fact it was made via LLM isn't the reason for the decline itself, why not just decline using what we have? Primefac (talk) 13:11, 17 November 2024 (UTC)[reply]
It tells submitters that they should not trust LLM content blindly. They may have not checked the draft for NPOV or tone issues because they thought LLMs were infallible. Ca13:20, 17 November 2024 (UTC)[reply]
Agreeing with this. (I am also of the opinion that LLM generated articles should be outright banned, but I may not be in the majority there, and besides, it'd be a hell of a thing to try and enforce) CoconutOctopustalk13:56, 17 November 2024 (UTC)[reply]
I'm also on the stricter side than current consensus. Lazy editors who trust software too much without proofreading and correcting its output are likely to create more problems down the road if not firmly dissuaded, and we should be openly unwelcoming to that behaviour. Also it's deeply unfair to write a prompt for an automated process to compose an article, then ask a human to spend more effort reviewing it than you spent thinking about it. Folly Mox (talk) 15:47, 17 November 2024 (UTC)[reply]
I think a specific "we see what you did there, and we don't like it" is helpful in this particular case, for two main reasons. One, most people using LLMs, obviously, have no idea how bad they are at doing the thing they want them to do, and a specific decline message could tell them that. Two, we don't actually want someone to waste their time trying to clean up and find sources for their garbage DraftGPT, do we? Personally, I'd prefer they just started over. -- asilvering (talk) 18:39, 17 November 2024 (UTC)[reply]
For a laugh I asked Chat GPT: "Please create a warning for a wikipedia editor who uses a Large Language model to generate articles, making sure that the understand that it is not appropriate. Then create a second wring against using LLMs in wikipedia talk page responses and conversations"
It replied:
Warning 1
Using Large Language Models to Generate Wikipedia Articles
Subject: Warning: Inappropriate Use of LLMs for Article Creation
Hello,
It has come to our attention that you have been using a Large Language Model (LLM) to generate Wikipedia articles. While LLMs can be helpful tools for research and drafting, their outputs often lack the rigor, neutrality, and verifiability required for Wikipedia content.
As per Wikipedia’s content policies, all articles must adhere to:
• Verifiability: Information must be supported by reliable, published sources.
• Neutral Point of View (NPOV): Content must be unbiased and factual.
• No Original Research: Contributions must be based on published information, not generated content.
LLMs often produce content that:
1. Lacks verifiable sources.
2. Introduces subtle bias or inaccuracies.
3. Contains synthesized information that may not reflect reliable sources.
This behavior undermines Wikipedia’s integrity and is considered inappropriate. If you continue using LLMs for article generation, further action may be taken, including editing restrictions or account sanctions.
Please ensure your contributions are original, well-researched, and comply with Wikipedia’s guidelines moving forward.
Thank you for your cooperation.
Warning 2
Using Large Language Models in Wikipedia Talk Page Responses
Subject: Warning: Inappropriate Use of LLMs on Talk Pages
Hello,
It has been observed that you may be using a Large Language Model (LLM) to generate responses on Wikipedia talk pages. While LLMs are capable of producing coherent text, their use in discussions is discouraged because:
1. Lack of Accountability: Wikipedia is built on collaborative, human input, and responses from LLMs lack personal accountability and intentionality.
2. Potential for Inaccuracy: LLM-generated responses may include subtle inaccuracies or misinformation.
3. Disruption of Collaboration: Talk pages are intended for genuine human dialogue and decision-making about article content.
Please refrain from using LLMs in talk page discussions. Instead, engage authentically with your own thoughts and words. If you need assistance in contributing effectively, refer to Wikipedia’s policies or seek help from experienced editors.
Continued misuse of LLMs may result in restrictions on your account or contributions.
For the record, I have found LLMs to be an extremely useful tool for taking a set of statements from sources and constructing them into a good start at a coherent and encyclopedic presentation of that content. Of course, content that is entirely generated by an LLM without using provided sources is going to be problematic, but we should not throw out the use of the tool as a tool in itself. BD2412T21:52, 17 November 2024 (UTC)[reply]
While I agree with the point that LLMs aren't banned and that "generated by LLM" by itself isn't a decline reason, it could be very handy to have a way to decline a draft for showing the problems usually associated with LLM content (promotional tone issues, generic speculative statements, "conclusion-like" paragraphs, essay-like wording trying to emphasize the subject's place in a broader context, etc.)These drafts are sadly way too typical to not have a specific decline reason. If the person used, say, ChatGPT, without knowing its limitations and what should be done to use AI-generated content constructively, then a generic "this article has tone issues" will be far less helpful than a "here is what you can do to improve raw LLM content". Chaotic Enby (talk · contribs) 23:02, 17 November 2024 (UTC)[reply]
Useful as a template especially when doing NPP, but I was thinking more of a decline reason (I'm loathe to put a template on a draft article as I'd like to think they wouldn't need any once published). CoconutOctopustalk10:56, 18 November 2024 (UTC)[reply]
Draft decline text (LLM)
Working on a draft decline message, would be happy to have feedback on it!
Your draft shows signs of having been generated by a large language model, such as ChatGPT. While the use of large language models isn't disallowed, their outputs usually have multiple issues that prevent them from meeting our guidelines on writing articles. These include:
Looks good to me - I'd maybe have something in there saying to please rewrite in your own words, just in case it isn't clear enough that they should be doing that. CoconutOctopustalk17:35, 18 November 2024 (UTC)[reply]
I disagree. An encyclopedia demonstrates what is believed to be established knowledge; this requires finding, reading, assessing, and analysing sources prior to writing a Wikipedia article. Encouraging editors not to learn this principle is anything but useful to this project. Content generated by LLMs is mostly useless for Wikipedia because LLMs are not knowledgeable. They excel at composing text that is sufficiently convincing to the average ignorant person if that text does not require any sort of factual accuarcy. For Wikipedia, compelling but erroneous content is extremely dangerous. Telling someone, who has already demonstrated a failure to understand what Wikipedia is, to rewrite unsensible LLM text is hopefully ignored by that addressee. Best, --Johannes (Talk) (Contribs) (Articles) 18:35, 18 November 2024 (UTC)[reply]
Just to clarify, this would be in conjunction to the final sentence that currently reads Furthermore, it is your responsibility to accurately verify LLM-generated content, as it will often contain hallucinations and fictious references. While it could be worded a bit more strongly, I believe this already tells editors that they should be verifying the factual accuracy of LLM-generated content, especially regarding sources. Chaotic Enby (talk · contribs) 18:38, 18 November 2024 (UTC)[reply]
Do we expect editors having to use LLMs to comprehend this? I'm drifting off course here, but I'm not sure whether you are aware that, to a certain group of people, using LLMs appears like a splendid way to circumvent the necessity to learn English. --Johannes (Talk) (Contribs) (Articles) 18:51, 18 November 2024 (UTC)[reply]
That is indeed a pretty good point. Maybe we could replace these two sentences by:
You should research and write content by yourself, and not trust what AI models write for you, as it is often made up. It is your responsibility to fully verify LLM-generated content that you introduce on Wikipedia, as it will often contain hallucinations and fictious references.
Consider deleting "While the use of large language models isn't disallowed". Even if it isn't technically disallowed, I think this sentence plants a seed in editor's heads that LLM use is OK, and giving this advice to a new editor is probably not a good idea. Consider also deleting the last sentence, "Furthermore, it is your responsibility to accurately verify LLM-generated content, as it will often contain hallucinations and fictious references.", for the same reason. This makes it sound like it's OK to have LLMs write articles as long as you double check their output. Which is not good advice at all. The basis of new articles should be reading and summarizing reliable sources, not asking ChatGPT to hallucinate an article.
Might as well add a new sentence to the end, too. Something like "Please address these issues. The best way to address these issues is usually to read reliable sources and summarize them, instead of using a large language model.".
So to summarize, I'd like to propose these modifications:
Your draft shows signs of having been generated by a large language model, such as ChatGPT. Their outputs usually have multiple issues that prevent them from meeting our guidelines on writing articles. These include:
Please address these issues. The best way to address these issues is usually to read reliable sources and summarize them, instead of using a large language model.
Thanks, that is indeed much better advice. Small detail, there is some repetition in Please address these issues. The best way to address these issues is.... Would The best way to do it is... work instead? Chaotic Enby (talk · contribs) 23:06, 18 November 2024 (UTC)[reply]
Feel free to put my version in a sandbox somewhere, link it here, and start modifying. I think we're at the point where folks being able to boldly edit the message and iterate on it would be helpful. –Novem Linguae (talk) 23:33, 18 November 2024 (UTC)[reply]
