Wikipedia talk:WikiProject AI Cleanup#AI-upscaling image cleanup template

Wikipedia:WikiProject AI Cleanup/AI images in non-AI contexts has been documenting reasons given for removing AI-generated images from Wikipedia articles, since 2023. Is there any reason to continue keeping track of this, now that WP:AIIMAGES has become policy? I assume the list page was created to help guide that eventual policy with organic examples from across Wikipedia, which would mean it was no longer really needed. Belbury (talk) 11:37, 12 May 2025 (UTC)

:Yep, most of them have been deleted, and "what to do" is much clearer with the policy. Borderline cases (which will be less frequent, but will certainly happen) can be discussed on this very noticeboard. Chaotic Enby (talk · contribs) 14:06, 12 May 2025 (UTC)

::Okay, I've marked the page as {{t|historical}}. Belbury (talk) 18:31, 2 June 2025 (UTC)

There is a request for cleanup assistance at {{slink|WP:ANI#Cleaning up after User:M1rrorCr0ss's mess}}, which involves over 2,000 edits that need to be reviewed for AI-generated content. Some details were mentioned in an earlier discussion, {{slink|WP:ANI#User:M1rrorCr0ss creating articles with fake sources, possibly with LLMs}}. — Newslinger talk 12:30, 21 May 2025 (UTC)

:I've (hopefully) deleted all articles I can find created by M1rrorCr0ss, but (a) I'm not absolutely sure I've got them all, and (b) there are still the huge number of redirects and an unknown amount of garbage content inserted into other, legitimate, articles. Are there any tools for digging this sort of thing out, to allow root-and-branch removal of contributions by an editor? — The Anome (talk) 11:09, 22 May 2025 (UTC)

::I don't know of any specific tool for that, but one could probably be coded using Wikipedia:WikiBlame to find the editor's additions. Chaotic Enby (talk · contribs) 11:20, 22 May 2025 (UTC)

:::[https://xtools.wmcloud.org/topedits/en.wikipedia.org/M1rrorCr0ss The Edit Counter can identify all pages with live edits by this user], but not if their content is still in those articles. –LaundryPizza03 (dc̄) 04:40, 24 May 2025 (UTC)

Would it be possible to create a bot that would check new articles, follow all embedded links, such as citation links, and attempt to fetch them? 404-ing and similar reference links are an obvious sign of lazy AI slop, and it would be easy to catch these early using this, and to tag articles for examination by editors. It could also try to check the linked references for at least some reseblance to the subject of the article: either through simple text comparison, or a ML method such as comparing embeddings (of which text comparison is a trivial example). It would obviously not detect sophisticated AI slop, but that's another issue entirely.

The obvious problem is the anti-crawler features of websites themselves that would tend to block accesses by the bot. Are there any services that can provide this kind of crawler access to third party sites in an ethical way, for example via a WMF-brokered use-whitelisted API obtained via an organization like Google, Cloudflare, Microsoft, Kagi ([https://kagi.com/bot]) or the Internet Archive who have generally unrestricted access to crawling (something like, say, Google's "Fetch as Google" service)? — The Anome (talk) 10:42, 23 May 2025 (UTC)

:See also this: https://news.ycombinator.com/item?id=23149841 While slow, the IA's fetch would be ideal for this purpose. Combined with a cache, it would be highly effective. It doesn't really matter if it takes several minutes to do a fetch, for the purposes of bots, which can take as long as they like. Because it would get a lot of hits, it would probably have to be a service agreeement with the IA to prevent it being rate-limited or blocked by them. The IA also seems to offer an API: https://archive.org/help/wayback_api.php — The Anome (talk) 11:24, 23 May 2025 (UTC)

:Some AI generated content possibly goes under the radar. So, this bot proposal is a good idea. But this will only be good for new articles, which needs to undergo patrolling, so there is already some human supervision. For AI editors expanding existing articles with fake references, bot would need to check every article that has seen a recent edit. —CX Zoom[he/him] ^{(let's talk • {C•X})} 12:49, 23 May 2025 (UTC)

::Absolutely. It will only catch the very dumbest AI slop content, but it appears that is currently low-hanging fruit, and still worth doing. I really like the idea of a content cache for already-fetched reference content; automated checking of references is a really promising research area, and one, I think, where using LLMs is entirely valid, if it is used with the correct threshold settings, so that it is more sceptical than the average human reviewer, and bad references can either be flagged as wholly bad (naive slop detection) or simply questionable (detecting either superior-quality slop, vandalism, or mediocre human contributions), and human review can then take over. — The Anome (talk) 13:30, 23 May 2025 (UTC)

:::I'll take the opportunity to point to #WP:UPSD Update above, in case {{ping|The Anome}} didn't see it. Headbomb {t · c · p · b} 13:45, 23 May 2025 (UTC)

::::Looks like it's been archived, but it's at Wikipedia talk:WikiProject AI Cleanup/Archive 2#WP:UPSD Update if anyone still wants it. Kudos to everyone working on it! Chaotic Enby (talk · contribs) 17:11, 30 May 2025 (UTC)

:::::link-dispenser.toolforge.org (a tool I wrote) also exists to check if a link is dead, it directly makes requests instead of routing through IA since IA heavily ratelimits Toolforge. Sohom (talk) 19:23, 30 May 2025 (UTC)

::::::Amazing! Chaotic Enby (talk · contribs) 20:43, 30 May 2025 (UTC)

Should this talkpage be considered the LLM noticeboard (perhaps adding a couple of redirects like WP:LLMN and Wikipedia:Large language models/Noticeboard?)? If not, should one be made? I wonder because I came across Zaida, Khyber Pakhtunkhwa and wanted someone more familiar with LLM to take a look, though I did find a maintenance template I added to the article. Gråbergs Gråa Sång (talk) 05:31, 24 May 2025 (UTC)

:I think it should. That seems like the intent of the WP:AINB shortcut, and there is precedent in designating a maintenance-oriented WikiProject talk page as a noticeboard: see Wikipedia talk:WikiProject Spam, which is listed on {{t|Noticeboard links}}. — Newslinger talk 06:19, 24 May 2025 (UTC)

:I agree with Newslinger, this has been de facto our LLM noticeboard, and it makes sense to have WP:LLMN and similar shortcuts redirect here. Chaotic Enby (talk · contribs) 12:35, 24 May 2025 (UTC)

:To facilitate searching for specific discussions in the archives, I suggest the active participants on this talk page should consider if it wants to keep project discussion separate from discussions of specific situations. isaacl (talk) 15:41, 24 May 2025 (UTC)

::That could also be a good alternative, assuming there are too many discussions and searching them ends up overwhelming. However, some discussions of specific situations can easily end up broadening in scope, so a separation between them might not always be practical. Chaotic Enby (talk · contribs) 15:46, 24 May 2025 (UTC)

::I do think that a separately maintained page will be better, because I can only see the issue grow in size in future. —CX Zoom[he/him] ^{(let's talk • {C•X})} 17:57, 24 May 2025 (UTC)

File:Symbol watching blue lashes high contrast.svg You are invited to join the discussion at Wikipedia:Village pump (WMF) § RfC: Adopting a community position on WMF AI development, which is within the scope of this WikiProject. Chaotic Enby (talk · contribs) 17:10, 30 May 2025 (UTC)

File:Symbol watching blue lashes high contrast.svg You are invited to join the discussion at Wikipedia:Village pump (technical) § Simple summaries: editor survey and 2-week mobile study, which is within the scope of this WikiProject. Chaotic Enby (talk · contribs) 15:23, 4 June 2025 (UTC)

All of the edits made by new user User:1january2000 the past few days and the fast rate at which they have been made considering the amount seem to be almost entirely A.I.-generated in volume, with many of the sources they've cited seeming to not actually exist, although referenced as if real. I am not sure what to do about this, but this seems like the right place to report it. Hellginner (talk) 17:38, 6 June 2025 (UTC)

:Looking at their last edit that added content (link), it very much appears to be full of made-up sources. The linked DOIs (doi:10.1016/j.spacepol.2044.11.002, doi:10.1103/PhysRevLett.280.234) and other URLs (https://unep.org/reports/climate-restoration-2870) don't exist at all, and are used to source content so speculative that it seems very unlikely that someone published these specific projections in serious journals.{{pb}}This, combined with the fact that they repeatedly write edits of thousands of bytes in a minute or so, makes it pretty obvious that AI editing is at play. Chaotic Enby (talk · contribs) 18:03, 6 June 2025 (UTC)

::yeah, this is probably AI, but much of it seems to fall afoul of WP:CRYSTAL regardless of the fictional sourcing, so I've reverted their changes. Psychastes (talk) 01:12, 7 June 2025 (UTC)

:My general approach with articles that are mostly or all LLM hallucinations, particularly if a chunk of references are clearly made up sources, is to tag them for speedy deletion as hoaxes with {{tl|db-hoax}}. As that template doesn't seem to have a comments or rationale field, I usually add in my analysis and rationale as an HTML comment too. Cheers, SunloungerFrog (talk) 18:29, 6 June 2025 (UTC)

::P.S. dealing with a similar case, user Vinizex94. Cheers, SunloungerFrog (talk) 18:31, 6 June 2025 (UTC)

:I've posted a warning message at {{slink|User talk:1january2000#June 2025}}, as {{np2|1january2000}} had not been warned yet. I've also invited them to participate in the discussion here. — Newslinger talk 19:02, 6 June 2025 (UTC)

:I have already taken care of a few edits by the same user on "Millennium celebrations" (section about Rio and South Georgia) which cited nonexistent sources from Folha de Spaulo, British Antarctic Survey among others. Ramkarlo82 (V • T • C) 01:25, 7 June 2025 (UTC)

A user on Talk:Bidirectional search alerted me to a problem with mass additions of content with hallucinated fake references by User:Noshin Nawal on bidirectional search. I have reverted the article to a version before the additions, and Noshin Nawal has not contributed to any other article, but I thought I'd leave this here in case it sounds familiar to anyone or might be helpful as a record of this action. —David Eppstein (talk) 22:06, 8 June 2025 (UTC)

Hi hi, the team behind Tone Check, a feature that will use AI prompt people adding promotional, derogatory, or otherwise subjective language to consider "neutralizing" the tone of what they are writing while they are in the editor, will be hosting a community consultation tomorrow on the Wikimedia Discord voice channels from 16:00 UTC to 17:00 UTC. Folks interested in listening in joining in, asking questions should join the Wikimedia Discord server and subscribe to [https://discord.gg/wikipedia?event=1380664800656363671 this event] Sohom (talk) 19:13, 9 June 2025 (UTC)

:Thanks a lot! I had the event noted already, but it's great if more people know about it. @Sohom Datta, should an alert also be sent at WP:VPT and/or WP:VPW? Chaotic Enby (talk · contribs) 19:32, 9 June 2025 (UTC)

::Notified! Sohom (talk) 21:20, 9 June 2025 (UTC)

I don't know where else to say this. I noticed [https://en.wikipedia.org/w/index.php?title=Aquatic_animal&diff=prev&oldid=1287564620 this revision] and this article, The Impact of Oil Spills on Aquatic Animal Life in the United States, which are clearly created with the assistance of a LLM.

I am a newcomer and I don't know how these are handled. What should be done about this? I genuinely don't think the article is a good fit for an encyclopedia, and checking/reworking everything that was included in the linked revision is a huge chore. I couldn't verify most of the sources used. I don't know if they're real, though I manage to find at least one of them. MeowsyCat99 (talk) 13:14, 13 June 2025 (UTC)

:Heavily LLM-generated and clearly not verified. I'm willing to put it up for AfD and advocate for TNT as I don't think attempting to salvage that level of generated content is worth any editor's time, not to mention other fundamental issues with the topic. Will wait a short time to see if any editors have a better suggestion. fifteen thousand two hundred twenty four (talk) 14:13, 13 June 2025 (UTC)

:My general approach now is to strip out made up sources and those that don't corroborate the sentence or paragraph they're attached to, and then send the article to draft with a reason of LLM-written text. I have also tried nominating for speedy deletion as {{tl|db-g3}} (hoax/vandalism) if it is particularly bad. In this case I'd probably try the draftify approach: I note that the creating editor is part of [https://dashboard.wikiedu.org/courses/University_of_New_Orleans/Information_Literacy_and_Scholarly_Discourse_(SPRING_2025_B_SESSION) this WikiEd course] so would probably notify the course leader / WikiEd person too. Cheers, SunloungerFrog (talk) 14:34, 13 June 2025 (UTC)

::[https://en.wikipedia.org/w/index.php?title=Special:Log&logid=170401816 Draftified]

::I would normally still advocate deletion to avoid other editors unknowingly getting caught up in LLM-cruft when trying to improve other's drafts, but will give draftifying a try this time.

::The creator has been warned with {{tl|uw-ai1}} and is now aware that using generated content like that is problematic, if they persist then I agree that contacting a course leader would be appropriate. fifteen thousand two hundred twenty four (talk) 02:52, 15 June 2025 (UTC)

[https://en.wikipedia.org/w/index.php?title=Guyana&curid=17238691&diff=1296376261&oldid=1294497399]

Not sure what to do. Doug Weller talk 16:50, 19 June 2025 (UTC)

:I've told the editor. Doug Weller talk 16:51, 19 June 2025 (UTC)

::Hi Doug,

:: FWIW, here are the three prompts I used from Gemini 2.5 Flash:

::1) Can you generate an updated economic summary using 2024 data for Guyana using the format below, and providing referenced sources for each data point that could be integrated into the Wikipedia page for it located at

::https://en.wikipedia.org/wiki/Guyana

::2) Can you also provide in Wikipedia format the list of references in your prior answer, also including verified working http links to webpages for each one?

::3) Can you

::1) find an alternative source than the website tradingeconomics.com for that reference, and if you cannot, remove that data and reference as it is blacklisted by Wikipedia

::2) and then provide a combination of the last two answers as a single body of Wikipedia text markup , modeled on the format below, but integrating the data you have just collated in the past two answers. Please double check that both the data and coding for Wikipedia markup are accurate.

::And then I made hand-tweaks of a few things that weren't perfect.

:: Is there a Wikipedia good-faith-AI crew collating efforts like this?

::It makes no sense to have the world's data centers regenerating the same kinds of outputs afresh when efforts could be strategically coordinated to flow the data to Wikipedia (among those inclined to do so).... Vikramsurya (talk) 17:02, 19 June 2025 (UTC)

:::The problem is this, from your edit summary {{tq|Data needs full verification but preliminary suggests it's accurate}}. You should only make edits that you have already fully verified are borne out by the sources, not just a vague suggestion that they're probably accurate. There are also three random inline citations on a line by themselves after the Imports bullet, and there's something wrong with the formatting of ref 57. Cheers, SunloungerFrog (talk) 17:25, 19 June 2025 (UTC)

:::PPP sources are broken, the sites list the data as being both for Guyana and Chad. Under "arable land" the hectare claim is not found in the source. Under "labor force participation" the rate in the source is 49.6%, not 56.4%. Under "industrial production" neither source mentions crude petroleum, gold, timber, or textiles.

:::The model's output can be characterized as "subtly wrong", this is par. fifteen thousand two hundred twenty four (talk) 19:15, 19 June 2025 (UTC)

::::AI hallucinating? Doug Weller talk 19:39, 19 June 2025 (UTC)

:::::Possibly some hallucination, but sourcing misattribution has certainly occurred, which can be viewed as better or worse. The arable land claim of 420,000 hectares (but not {{tq|"more than"}}) is the exact figure in Wolfram's database, but the prompt requested {{tq|"working http links to webpages"}}, so the model's pattern contained a link, even if wrong. fifteen thousand two hundred twenty four (talk) 04:39, 20 June 2025 (UTC)

::::::Misattribution and hallucination are really the same issue, the AI is finding words and numbers that fit the pattern it develops. CMD (talk) 05:31, 20 June 2025 (UTC)

:::I have a question - when did you think the verification by other editors would occur? If I was watching the page and started to check and found more than a couple of errors, I would just revert the whole edit with a request not to submit error-strewn material. Why? Because I would judge that the edit overall could not be trusted if there were already this many faults and I wasn't going to waste my time looking further. This is something that happens all the time: we are all volunteers who shouldn't be making work for each other like this. That doesn't mean using an LLM is bad. It's saved you time doing some of the formatting. That frees you up to do what the LLMs are bad at, which is fine-grained fact-checking of reliable sources. OsFish (talk) 05:44, 20 June 2025 (UTC)

I'm not super familiar with the process here, but Royal Gardens of Monza seems like it might be AI generated to me - two of the books it cites have ISBNs with invalid checksums, the third doesn't seem to resolve to an actual book anyways, it cites dead URLs despite an access date of yesterday, and uses some invalid formatting in the "Design and features" heading. The author has also had a draft declined at AFC for being LLM-generated before. ScalarFactor (talk) 23:07, 21 June 2025 (UTC)

:You are correct. I've [https://en.wikipedia.org/w/index.php?title=Draft:Royal_Gardens_of_Monza&diff=prev&oldid=1296749320 draftified] and [https://en.wikipedia.org/w/index.php?title=Draft:Royal_Gardens_of_Monza&diff=prev&oldid=1296749478 tagged] the article, left notices on the [https://en.wikipedia.org/w/index.php?title=Draft_talk:Royal_Gardens_of_Monza&diff=prev&oldid=1296749816 draft] and [https://en.wikipedia.org/w/index.php?title=User_talk:Fazzoacqua100&diff=prev&oldid=1296751226 creator's] talk pages, and [https://en.wikipedia.org/w/index.php?title=User_talk:Bkissin&diff=prev&oldid=1296750410 notified] the editor who accepted the draft at AfC. I think Fazzoacqua100's other articles should be reviewed for similar issues. fifteen thousand two hundred twenty four (talk) 01:22, 22 June 2025 (UTC)

::Their other submissions and drafts have now been reviewed, draftified, and had notices posted where appropriate. Thank you @ScalarFactor for posting here. fifteen thousand two hundred twenty four (talk) 04:40, 22 June 2025 (UTC)

:::No problem - thanks for dealing with the cleanup. ScalarFactor (talk) 05:15, 22 June 2025 (UTC)