Jump to content

User talk:Polygnotus

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Eine Treppe

DuplicateReferences question

[edit]

Hi Polygnotus, is it possible to override the automatic edit summary when using your script? I was thinking of using something like: Tagged duplicate citations using DuplicateReferences Nobody (talk) 12:37, 28 May 2025 (UTC)[reply]

@1AmNobody24: Sure, see here. Polygnotus (talk) 12:46, 28 May 2025 (UTC)[reply]
Thank you! Nobody (talk) 12:48, 28 May 2025 (UTC)[reply]
Not sure if you have a bugtracker, but it just did this. Nobody (talk) 13:29, 28 May 2025 (UTC)[reply]
Hm, I should look into the morebits template insertion logic, one sec. Polygnotus (talk) 13:39, 28 May 2025 (UTC)[reply]
@1AmNobody24 I am now using the Wikipedia:Morebits template insertion logic, which should work now. Polygnotus (talk) 13:47, 28 May 2025 (UTC)[reply]
Thanks for the quick fix. Nobody (talk) 13:48, 28 May 2025 (UTC)[reply]
Yeah it was already on the todolist but I hadn't gotten around to it yet. Polygnotus (talk) 13:50, 28 May 2025 (UTC)[reply]
Continued at MediaWiki_talk:Gadget-morebits.js#Morebits_and_MOS:ORDER. Polygnotus (talk) 21:50, 28 May 2025 (UTC)[reply]

Tech News: 2025-23

[edit]

MediaWiki message delivery 23:52, 2 June 2025 (UTC)[reply]

On VPT

[edit]

Can I ask you take a step back and consider not WP:BADGERING Olga  ? Also consider toning down on asking rhetorical questions or by default assuming that consensus will be overridden by surveys in this discussion (like you did with respect to MMiller).

Note that these folks are real people with real jobs that typically include a large number of disparate projects, they are typically not single threading on a specific thing (and thus are not required to answer every single pointy question thrown at them by the community and neither should you try to get them to do it). Sohom (talk) 13:38, 5 June 2025 (UTC)[reply]

@Sohom Datta Please read the page you linked to. I do not think it is polite to falsely accuse someone. You can disagree with me, that is fine. But pretending that I am doing things I am clearly not doing, just because you have a different position in a discussion is weak sauce.
Note that I am also a real person, and your approach of not engaging in discussion but falsely accusing me of something I did not annoys me. Please stay off my talkpage if you have nothing to offer but negativity. Thank you. Unlike you I am actually working towards a solution and helping the WMF. Polygnotus (talk) 13:44, 5 June 2025 (UTC)[reply]
@Sohom Datta: You claim to be an expert JavaScript programmer on your userpage. If you want to contribute something meaningful and constructive, I have this but I am not sure what the best way is to have it occupy the side of the screen. Normally that wouldn't be difficult but for some stupid reason the different themes Wikipedia uses mess things up. Claude couldn't figure it out and I am currently too sick for debugging. The idea is to use it to prove to the Wikipedia community that not all AI is bad. You can look at User:Polygnotus/Scripts/Claude3.js. I use Vector 2010 but I tried to get it to work in modern Vector. Figuring out what to do is not hard: 1) WMF should acknowledge the problems and show that it understands 2) Someone should show the community that not all AI is bad. You can help me with both. Polygnotus (talk) 13:51, 5 June 2025 (UTC)[reply]
So, I realize my initial message was a bit hasty and poorly worded. I wanted to dissuade you from making comments like Probably a bad idea to respond only to 1 comment, the only comment that is not critical but asks about the underlying stuff. We are kinda waiting for a response to the other stuff. since it makes people feel like you are demanding that folks answer the hard questions, even if you don't want it to sound that way. I can however, understand that this is a fairly emotionally charged situation and both of us worded our responses poorly.
Regarding the rest, User:Sohom Datta/claude.js should contain a fixed version of the script that you showed me. For my part, while I might seem relatively inactive onwiki, I've been doing my part as a member of the PTAC, I've raised this internally (on the PTAC slack) and pointed out that the English Wikipedia is perceives large-language model generated text negatively. I've also asked WMF folks to reconsider it's internal guidelines for what can/should be A/B tested without community approval (since I see this as being the main reason/loophole why this made it to a enwiki deployment). I also plan on doing some ground work trying to understand how/why the WMF made the decision of proposing this feature and see if I can provide feedback about the workstream. Sohom (talk) 00:13, 6 June 2025 (UTC)[reply]
@Sohom Datta Thank you! One of the problems here is that, if you try to push a discussion in a direction, people link to WP:BADGER which is written for bad people with bad intentions, and not people who are, let's say, perhaps a bit overly enthusiastic. Saying "Oi, slow down" works better. Badgering is demanding that people restate their position, refusing to accept the consensus, IDHT behaviour. I didn't do any of that, and I very much agree with the consensus.
I know I can convince the community that we shouldn't throw out the baby with the bathwater. I want to continue using AI, and I want the WMF to continue experimenting, while also being very very clear that some things go against our values.
I've been thinking about voluntarily tagging my own AI-assisted edits with a special (hash)tag so that people can see that the fear of AI (in a supporting role, while the human makes the decisions and holds the responsibility) is overblown. I don't use the generated text on Wikipedia, I just have it generate instructions that I can follow (like "fix this typo").
Stuff like this is a major setback because the community loses confidence that the WMF respects the community and understands its own role. Creating a simple script like the one I showed you and posting some screenshots on a public place is a quick win. I want to show the WMF what acceptable and responsible AI use looks like and show the community that they have nothing to fear from responsible AI use. We can't just shout "bad dog" at the WMF and expect them to do better next time. Anthropic's Sonnet 4 and Opus 4 work really well in my experience. It may be a good idea to also make a ChatGPT-version. In the future, maybe we can have a script like this with a model hosted by the WMF.
I do think that calling out manipulative tricks is very important. There is probably a better term for this I am unaware of, but let's call it "politician-speak". I think the WMF uses politician-speak to protect itself and does not realize that it is hurting instead of helping. We need to be able to criticize undesirable behaviour without wrapping it in compliments. Way too often people hide behind the trick "oh, you criticized something I said, then you must be a one-dimensional villain/troll who is just plain evil and has nothing of value to add" instead of realizing that their behaviour is not who they are as a person and that they can't grow if no one points out when they mess up. It may have something to do with the world getting more polarized/people being less relaxed but I am no expert in that field. I am on the other end of the spectrum in the sense that I rarely agree with myself, and can write books about the flaws in my own reasoning.
I may want to get community consensus for my AI-assisted typo fix tool at some point, and people will object to it if they think all AI is evil and wrong.
Thanks for your help. I think we need to turn this debacle into something positive, somehow. If you need help with PTAC stuff or other stuff, you know where to find me. Polygnotus (talk) 01:31, 6 June 2025 (UTC)[reply]

Todo

[edit]

Stuff that should probably happen:

  • use OOUI buttons (but they use too much space, because they are fullwidth)
  • think about accessibility
  • make OpenAI version(?)
  • add warning about AI imperfections
  • preserve scroll position when navigating to new page
  • currently it opens on top of the page, and when you drag the left border the article width changes. It should never be on top, and always next to article content
  • make button to delete API key from localStorage() instead of pointless "clear results" button
  • key should probably be in sessionStorage not localStorage
  • improve prompt to not include summary at the end The article is generally well-written
  • improve prompt to ensure it focuses on areas in which it gives good advice, and ignores areas about which it gives bad advice. Should probably run it on ~100 pages and check what is the worst advice and tune the prompt based on that. Special:NewPages >10k b?
  • horribly ugly in light mode
  • ability to minimize and restore?
  • it is relatively pointless on pages with very little text/pages that consist mostly on tables with scores. Perhaps have a way to detect that it shouldn't even bother trying?
  • how to deal with very very long pages? I got Quarry 91712 and Quarry 91714. Test on https://en.wikipedia.org/wiki/List_of_chiropterans
  • test with other themes
  • when its working on the next article it could hide the output from the previous one
  • probably possible to get how much money is left on the API key
  • improve error handling
  • strip ref and exlink sections (and infobox?) before sending wikicode
  • in system prompt, explain that things get repeated in the lead and that that is nothing to worry about
  • i18n

Administrators' newsletter – June 2025

[edit]

News and updates for administrators from the past month (May 2025).

Administrator changes

removed

Interface administrator changes

added 0xDeadbeef

CheckUser changes

readded L235

Oversight changes

readded L235

Guideline and policy news

  • An RfC is open to determine whether the English Wikipedia community should adopt a position on AI development by the WMF and its affiliates.

Technical news

Arbitration

  • An arbitration case named Indian military history has been opened. Evidence submissions for this case close on 8 June.

Miscellaneous


A free Gemini version of your Claude6.js proofreader

[edit]

Greetings Polygnotus! I'm a huge fan of your work on User:Polygnotus/Scripts/Claude6.js. Anthropic has to charge a lot of money but Google offers their slightly better (at text in LMArena, anyway) gemini-2.5-flash-preview-05-20 model for free, so I asked it to convert your code to User:Cramulator/GeminiProofreader.js which I have tested and seems to work. There is a hyperlink to Google's API key generation page in the key entry dialog box which anyone logged in to Google even without a Google Cloud account can use to generate a free tier key. The only technical change I had to make was to increase maxOutputTokens well above the default for when it had to say a lot, because unlike Claude, Gemini returns no completion at MAX_TOKENS instead of truncating, nor does it prioritize generation to fit inside the limit in the first place. I also made it prepend the article title to the LLM results since the output persists (which I think is a good thing, while you work on an initial article; but this probably could have been a prompt change.)

I also enabled Gemini's googleSearch and urlContext tools, so presumably it should be using its "grounding" capability to search for facts and sources, and actually examine linked sources (including PDFs etc.) when prudent, but I haven't tested either of those yet because I'm not sure how I could set up conclusive tests -- maybe you can think of something?

...Maybe we should put something in the Gemini prompt to use web search to check important facts and figures, and browse sources to confirm important or conflicting statements? Cramulator (talk) 06:15, 9 June 2025 (UTC)[reply]

@Cramulator Check out User:Phlsph7/SourceVerificationAIAssistant.js and T360489. More detailed response follows. Polygnotus (talk) 07:54, 9 June 2025 (UTC)[reply]
@Cramulator Ooh that is excellent work, thank you! We may need to combine them at some point and add OpenAI. AI bashing is pretty popular round these parts, but Wikipedians should be able to deal with unreliable sources. I'll have to make something to compare them so we can judge who is best. I prefer Claude for coding tasks at the moment, but being free is a very important advantage of course! When I have more time I'll try to make some minor improvements. Thanks again! Polygnotus (talk) 00:21, 11 June 2025 (UTC)[reply]
You're very welcome! I did very little compared to your 6 revisions. I agree the Claudes version 4 are better for code than competitors (when the much smaller context window doesn't get in the way), but Anthropic clearly didn't optimize them for prose over previous versions, while Gemini Flash 2.5 has three other advantages that I believe the Claude API lacks:
  1. Web search: Gemini will try to use Google Search with its "grounding" approach which I believe looks at low confidence token output behind the scenes and gets more aggressive with searching when it thinks it might be hallucinating. Claude 4 via the API won't search the web without "extended thinking" which is slow, expensive, and possibly overkill for proofreading.
  2. PDF browsing: Claude can read PDF files, but by default it refuses to. For example https://s24.q4cdn.com/216390268/files/doc_downloads/test.pdf which Claude says is forbidden by robots.txt, but there is no robots.txt or exclusionary headers for it, so.... I suggest many if not most of the PDF sources we link to in articlespace lack affirmative permissions from a robots.txt file. Also, when you turn on Gemini's urlContext tool as I did, I have confirmed it tries to load in as many links as it can, and keeps them when they fit in its million token window.
  3. Cost: Gemini Flash has a very generous free tier, while Anthropic and OpenAI have less of a slush fund to dangle "$0" in front of potential APIs users, so I expect Google will continue to try to bludgeon them this way. We definitely want to prioritize this for editors, not just to reduce adoption friction and keep them from having to open their wallets, but also while I know very little about userscripts and gadgets, I know security issues are not unheard of, and I would hate to see editors getting defrauded somehow. Gemini free tier API keys just rate limit, without incurring payments.
If I were going to offer just one point of constructive criticism, I suggest the three vertical buttons could be improved to be horizontal when width allows. Cramulator (talk) 02:58, 11 June 2025 (UTC)[reply]
P.S. Please feel free to fork User:Cramulator/GeminiProofreader.js back into your /scripts/ in which case I will make it a redirect to your version, or just edit it in my userspace as you see fit. Cramulator (talk) 03:07, 11 June 2025 (UTC)[reply]

Reporting DetectPromo false-positive

[edit]

2009_UK_&_Ireland_Greyhound_Racing_Year "significant" in phrase "significant changes" is not promotional CreatorTheWikipedian2009 (talk) 17:06, 19 June 2025 (UTC)[reply]

Reporting DetectPromo false-positive

[edit]

Khartoum_(2025_film) "premiere" in phrase ""Premiere" should not be included, as it often refers to the first showing of a film or theatre play." is not promotional Munfarid1 (talk) 10:22, 2 July 2025 (UTC)[reply]

Live from reddit

[edit]

https://www.reddit.com/r/Palestinian_Violence/comments/1ltjre6/this_aggressive_antijewish_user_dares_remove_the/. Looks like it’s your time to shine. Insanityclown1 (talk) 03:25, 7 July 2025 (UTC)[reply]

@Insanityclown1 Adorable! They posted it on no less than 4 subreddits. I assume the lynchmob will be here any minute. I'll go brew some coffee for them. Polygnotus (talk) 03:30, 7 July 2025 (UTC)[reply]
Signatures are 50 bucks. https://www.reddit.com/r/Jewpiter/comments/1lstfbg/how_dare_you_call_us_out_huh/ Polygnotus (talk) 03:56, 7 July 2025 (UTC)[reply]
https://www.reddit.com/r/FightAntisemitism/comments/1lt3tor/this_is_the_user_who_aggressively_deletes_content/ Polygnotus (talk) 11:02, 7 July 2025 (UTC)[reply]

Re: fame and antisemitic stuff

[edit]

That's ok, I've been famous for a long time, as a former admin and bureaucrat who has gotten into hot water more than once. Someone once found my personal website and a stand-up comedy routine that I did at an open mic night and posted it on Digg (not that it was well-hidden). They took it down when I asked nicely, I believe it is impossible to find now, but if you find it let me know 'cause I long since lost the video. Anyway, a word of unsolicited advice, a lot of people that I know are pretty sensitive about these issues (antisemitism on Wikipedia, and so on) especially what with what is going on in American politics right now (e.g. Trump using the word shylock, various horrific events of antisemitism in the country that I won't get into). I don't want to get into it any more than that, but my advice would be to try to tread a little lighter and consider being a bit more empathetic in terms of the impact your comments and actions may have. It is possible to both fix the coverage and articles and also be a bit more sensitive and respectful toward opposing viewpoints. I say this as someone who has been a hothead many times and I know that you are in good faith trying to improve Wikipedia. Andre🚐 03:55, 7 July 2025 (UTC)[reply]

@AndreJustAndre Thanks! Yeah, they don't see it when I protect Talia Lavin, or fix typos on every single Rote Kapelle article written by user Scope creep. With the goodfaith people it usually works out in the end, we follow BRD and come to a consensus or agree to disagree, but for badfaith people it is always easy to pretend you are an evil %insert bad group here%-ist. As someone who has spent time online I have been called a nazi, commie, socialist, n-word, racist, antisemite, islamist and the like. Of course I make it worse by being against both sides in the I/P conflict. I think the civilians there deserve so much better. It would be much easier to support one side and hate the other. Polygnotus (talk) 04:20, 7 July 2025 (UTC)[reply]
Echoing @Polygnotus. I don’t like bullies, ethnicity or creed has nothing to do with it. It’s why I decided to pursue a career in law was so I could advocate for people weren’t as lucky as I was. Insanityclown1 (talk) 07:29, 7 July 2025 (UTC)[reply]
@Insanityclown1 I do recommend simple.wiki it is like conservapedia:
Hamas (Arabic: حَمَاسُ, romanized: Ḥamās), an acronym of its official name, Harakat al-Muqawama al-Islamiyya (Arabic: حَرَكَةُ الْمُقَاوَمَةِ الِاسْلَامِيَّةِ, lit. "Islamic Resistance Movement"), is a Palestinian Sunni Islamist political movement in Gaza, with a history of terrorism and antisemitism despite its relative popularity in the Middle East and Western academia.[4] Polygnotus (talk) 09:25, 7 July 2025 (UTC)[reply]
He turned the entire article simple:Neo-Nazism into one about Nation of Islam and Islamism. Polygnotus (talk) 09:42, 7 July 2025 (UTC)[reply]
84 attempts to log into my account on meta. If they want to bruteforce it they'll need quite a bit of patience. https://irontechsecurity.com/wp-content/uploads/2021/07/Screen-Shot-2021-07-28-at-10.49.43-AM.png Pretty sure this is a false flag operation designed to make 'em look bad. Polygnotus (talk) 14:35, 7 July 2025 (UTC)[reply]

Contest invite list

[edit]

Where was the most recent filtered list you started compiling in A-Z again? We'll need to work a lot on that after the contest in time for the next contest, likely in the mid autumn. Some way of filtering it to the most active members of WikiProjects and content producers across the site would be ideal, though in some ways contests like this have a lot of potential in motivating people who might otherwise not edit much, that has to be taken into account.♦ Dr. Blofeld 12:17, 8 July 2025 (UTC)[reply]

@Dr. Blofeld
The list is at User:Polygnotus/Data/BloActive
Polygnotus (talk) 14:16, 8 July 2025 (UTC)[reply]
Thanks. Can I move it to User:Dr. Blofeld/Contest invitation list? I'll need your help building a filtered list over the next couple of months. I'm sure I'll gradually add a few manually too. ♦ Dr. Blofeld 14:32, 8 July 2025 (UTC)[reply]
@Dr. Blofeld Sure, I moved it there. Polygnotus (talk) 14:39, 8 July 2025 (UTC)[reply]
Thanks. I suppose the first step would be to add a list of participants in former contests and challenges not on that list, which can be accessed in Template:The 100,000 Challenge. Then, given the emphasis on Science, Technology, Engineering, Maths, Medicine, Business, Geography and architecture, to find active members from all WikiProjects or associated articles in those realms and add them to the list. No rush as there's plenty of time over late summer but would be great if you could help and not too much trouble. ♦ Dr. Blofeld 14:53, 8 July 2025 (UTC)[reply]
@Dr. Blofeld Didn't that one dude have an idea on how to use quarry? You should probably contact them. Polygnotus (talk) 14:55, 8 July 2025 (UTC)[reply]
Not sure. I can find people manually, but would be great if you or somebody could use something which is quicker and more efficient and build it faster. ♦ Dr. Blofeld 14:57, 8 July 2025 (UTC)[reply]
It was @Alien333: see the link. Polygnotus (talk) 15:00, 8 July 2025 (UTC)[reply]
(Just noting: my query was a bit messy, I'd rather recommend that one. To get a new list, fork the query and change the 'Science'.) — Alien  3
3 3
15:05, 8 July 2025 (UTC)[reply]
@Dr. Blofeld You don't have to find people manually, Alien333 provided a nice little query for you. Then you can use the scripts above to filter out blocked and inactive users and sort the list by editcount. Polygnotus (talk) 15:09, 8 July 2025 (UTC)[reply]
Can it be used for each branch of the science projects too, geography and architecture?♦ Dr. Blofeld 15:36, 8 July 2025 (UTC)[reply]
The problem is a lot of those on that list aren't real science editors and are maintenance people like Ser Amantio and Liz. ♦ Dr. Blofeld 15:37, 8 July 2025 (UTC)[reply]
I think Liz will be near the top of any list you generate. Liz does roughly 50% of the work round here. Entering "Liz" into XTools makes it crash. Can it be used for each branch of the science projects too, geography and architecture? Alien333 will know. The problem is a lot of those on that list aren't real science editors and are maintenance people like Ser Amantio and Liz Maybe Alien333 can come up with a way to filter people who made articles in a specific category? Just spitballing. Polygnotus (talk) 15:51, 8 July 2025 (UTC)[reply]
Entering "Liz" into XTools makes it crash. Well, to be exact, we choose to not do such expensive calculations.
Can it be used for each branch of the science projects too, geography and architecture? It can, as they're given in the assessments templates. But they won't be found if you search for science. There is a technological solution to denote parent projects, but the only enwiki projects that use it also add the parent project directly (so there's not much point adding that to the query).
The problem is a lot of those on that list aren't real science editors and are maintenance people I'm afraid that there's no programmatic way of differentiating a content edit and non-content edit (though there's been a lot of discussion around it).
Come up with a way to filter people who made articles in a specific category Depends what you mean by "made". If it's "made" as in first edit, then should be fairly straightforward; for most other senses, it'll be near-impossible (in reasonable time at any rate). — Alien  3
3 3
16:15, 8 July 2025 (UTC)[reply]
quarry:query/95313 is with only page creations. It is quite slower, especially for large projects, because we aren't using an index anymore. However, to the best of my knowledge (which granted is quite slight), there are no indexes on revision that include rev_parent_id. — Alien  3
3 3
16:37, 8 July 2025 (UTC)[reply]
@Alien333 What do you look at to figure out something like that? This? Polygnotus (talk) 16:39, 8 July 2025 (UTC)[reply]
Indexes are given in tables-generated.sql. Most other stuff is on the individual tables' pages (eg mw:manual:page table or mw:Extension:PageAssessments/page assessments table). — Alien  3
3 3
16:46, 8 July 2025 (UTC)[reply]
I think finding editors who created/significantly expanded articles in these fields rather than sheer edit count would be the way to go if it can be done. Including editors who wrote GAs and FAs.♦ Dr. Blofeld 16:49, 8 July 2025 (UTC)[reply]
Creation is simple. But again, knowing who wrote an article or significantly expanded it is AFAIK not doable on a large scale. The best tool I know for that is mw:WikiWho (enwiki API), but it relies (as I think would any tool that does this) on analysing the whole revision history; which we can't do for thousands of articles.
A possible approximation could be number of edits to FAs and GAs of a certain WikiProject in the last year (though we can't check when exactly it became a GA/FA; assessment changes are not logged). — Alien  3
3 3
16:58, 8 July 2025 (UTC)[reply]
God disagrees. It took him days! (allegedly) Polygnotus (talk) 17:00, 8 July 2025 (UTC)[reply]
Edits to GAs and FAs and creations would be a start. A list of science and STEMM editors with such articles would be good. ♦ Dr. Blofeld 17:44, 8 July 2025 (UTC)[reply]
https://en.wikipedia.org/wiki/Wikipedia:Village_pump_(technical)/Archive_220#List_of_Wikipedians_by_country_project Polygnotus (talk) 14:57, 8 July 2025 (UTC)[reply]

@Dr. Blofeld: Here it is: quarry:query/95322. To be precise, it's the 100 editors with the most edits in GA/FAs of interest to the science wikiproject and/or its child projects. As a bonus, I also excluded bots directly in the query. I've put the results at Special:PermaLink/1299489775. — Alien  3
3 3
18:32, 8 July 2025 (UTC)[reply]

That was the GA/FA edits in science &co list; I have also made the creations in science &co list. It's from quarry:query/95313 and I put the results at Special:PermaLink/1299492145. For the fun of it, I also made the list of GA/FA creations in science &co: Special:PermaLink/1299492308 (there are, understandably, only 32 people on that list).
Note: many editors appear in two of the lists. — Alien  3
3 3
18:53, 8 July 2025 (UTC)[reply]
Nice one, just updated it with the first list from your sandbox and hopefully removed the duplicates. The two of you feel free to add names to that list you find including from the second one. We want good contributors from all projects really, but STEM and Geography first place to start. Perhaps try Medicine next? Plenty of time, no rush! ♦ Dr. Blofeld 19:16, 8 July 2025 (UTC)[reply]

Talkback demo

[edit]
Hello, Polygnotus. You have new messages at Template talk:Talkback.
Message added 09:54, 12 July 2025 (UTC). You can remove this notice at any time by removing the {{Talkback}} or {{Tb}} template.

Current contest list.

[edit]

Hi, can you do two things. Wikipedia:The World Destubathon/Participants. Use that to generate a new full list with just User talk:Xxxx on each line for messaging, I should really notify all that the contest now ends on 16th. And two, add the list to my current A-Z list we worked on the other day in alpha order but remove duplicates? I think anybody who showed an interest automatically qualifies for notifying next time. ♦ Dr. Blofeld 16:44, 12 July 2025 (UTC)[reply]

@Dr. Blofeld See User talk:Dr. Blofeld/list. The FilterInactiveOrBlocked script handles deduplication. Polygnotus (talk) 16:16, 13 July 2025 (UTC)[reply]
Thanks, add them to my list too. ♦ Dr. Blofeld 16:47, 13 July 2025 (UTC)[reply]

Detecting AI/LLM edits

[edit]

Hello Polygnotus,

I often see you placing llm tags on articles and I would like to know how you come to such conclusions primarily because I want to do such research myself (If you don't mind, of course). I see more and more articles being infused with machine text, and in many cases it could be problematic.

Well regards, Plasticwonder (talk) 18:33, 12 July 2025 (UTC)[reply]

@Plasticwonder In some cases it is pretty obvious. Especially if they use somewhat plausible sounding sources which do not exist from websites that do not exist. There are also stylistic things like the rule of 3s and em dashes and all that. But I think one of the most recent article I tagged with {{LLM}} was Muriel Bowser and you can see that it says ?utm_source=chatgpt.com (see UTM parameters). See also Wikipedia:AutoWikiBrowser/Tasks#Tag_LLM_generated_stuff. Polygnotus (talk) 16:07, 13 July 2025 (UTC)[reply]
Thank you so much for this! Plasticwonder (talk) 00:50, 15 July 2025 (UTC)[reply]
@Plasticwonder See also Wikipedia:Signs of AI writing. Polygnotus (talk) 20:12, 21 July 2025 (UTC)[reply]
Ok, will do. Thanks! Plasticwonder (talk) 20:25, 21 July 2025 (UTC)[reply]

Wikidata weekly summary #688

[edit]

Tech News: 2025-29

[edit]

MediaWiki message delivery 20:06, 14 July 2025 (UTC)[reply]

Wikimedia Foundation Bulletin 2025 Issue 13

[edit]


MediaWiki message delivery 18:54, 15 July 2025 (UTC)[reply]

Feedback request: All RFCs request for comment

[edit]

Your feedback is requested at Talk:Ellie Reeves on a "All RFCs" request for comment. Thank you for helping out!
You were randomly selected to receive this invitation from the list of Feedback Request Service subscribers. If you'd like not to receive these messages any more, you can opt out at any time by removing your name.

(trialing replacing Yapperbot) SodiumBot (talk|botop) 01:31, 17 July 2025 (UTC)[reply]

[edit]

Hallo, When you retargetted the redirect Prof you don't seem to have added a hatnote at the new target to redirect readers to the previous target. (OK, it's all now up in the air with the redirect being under discussion at RfD, but the point is still valid.) Please remember to do so, for the sake of our readers. Thanks. PamD 16:26, 17 July 2025 (UTC)[reply]

@PamD I basically never do anything with redirects so I didn't know that was required. The next time I'll touch a redirect is probably 5 years from now so it is likely I will have forgotten by then. Thanks! Polygnotus (talk) 16:35, 17 July 2025 (UTC)[reply]
@PamD Also that sounds like a task that is pretty easy to automate. Maybe post on WP:BOTTASK. Polygnotus (talk) 16:48, 17 July 2025 (UTC)[reply]
Not that simple, there are lots of variations. But the bottom line is: Before this edit, a reader typing "Prof" was led to Professor, so we must make sure that they are still offered a route to get to that page even if they go via the rapper's page. Just think of the reader, and help them, even if they don't share your musical interests! PamD 22:02, 17 July 2025 (UTC)[reply]
@PamD Did you check the clickstream data? I looked at WikiNav but it says An error occurred while fetching data for the current title. Try another one.. Wikipedians have all these ideas about how our readers should use Wikipedia, but I am pretty sure they don't. The rapper has only 82 inbound links, yet gets almost the same amount of pageviews as Professor, a page with over 19k inbound links (most of which should be removed per WP:OVERLINK). It is possible but rather unlikely that someone is looking for a capable but not exceptional cricketer from the 40s, or a tld that is very rarely used. I think the principle of least astonishment applies here. Polygnotus (talk) 00:22, 18 July 2025 (UTC)[reply]
I'm not arguing about where the redirect Prof should go to: that's under discussion at RfD. What I'm saying is:
  • There was for many years a redirect from Prof to Professor, because someone thought it would help some readers
  • You retargetted it to the rapper because you thought it would help more readers
  • I suggest you should have helped the first lot by adding a hatnote "For the academic title see Professor" or similar, at the top of the rapper's page.
  • Actually, on further thought, you should probably have made it a formal Move discussion, to propose moving the rapper to an undisambiguated "Prof", because retargetting the redirect implies that he is the Primary Topic so should be at the base name.
PamD 05:16, 18 July 2025 (UTC)[reply]
@PamD That again sounds like a task that is automatable. Perhaps a bit more difficult, but surely doable. So I think that the solution is some code that figures out what the primary topic should be, and then deals with redirects and {{For}} and the like. Why waste valuable human time on such a task? If we have clickstream and pageview data some simple code is far more reliable and predictable than humans can ever be. If a page is incorrectly redirecting, like it was, then my edit was a clear improvement. Sure, it could've been even better by deleting the redirect and moving the article, but perfect is the enemy of good and I am no admin. So with the benefit of hindsight we can say Ohconfucius' edits were wrong, but unless Ohconfucius is a timetraveling wizard they probably couldn't reliably predict the future. Before this edit, a reader typing "Prof" was led to Professor, so we must make sure that they are still offered a route to get to that page I don't see the causal link, unless you think people type prof to find the article Professor over and over again, which seems unlikely. And of course people who want to learn about the concept of a "professor" are very unlikely to type the first 4 letters and then bash Enter and expect to get there. Usually you type the entire word, or you pick the correct option from the list that is presented to you while you type. Ideally we'd find a privacy-respecting way to collect data such as "time spent on page" and "lines scrolled", that could be used to improve the code. Polygnotus (talk) 05:30, 18 July 2025 (UTC)[reply]

Wikidata weekly summary #689

[edit]

Tech News: 2025-30

[edit]

MediaWiki message delivery 23:39, 21 July 2025 (UTC)[reply]

Wikidata weekly summary #690

[edit]

Tech News: 2025-31

[edit]

MediaWiki message delivery 00:23, 29 July 2025 (UTC)[reply]

Wikimedia Foundation Bulletin 2025 Issue 14

[edit]


MediaWiki message delivery 21:02, 29 July 2025 (UTC)[reply]

Reporting DetectPromo false-positive

[edit]

Ertuğrul "more than" in phrase "more than" is not promotional Chronos.Zx (talk) 06:07, 31 July 2025 (UTC)[reply]

Reporting DetectPromo false-positive

[edit]

Gupta_Empire "comprehensive" in phrase "Comprehensive, more than" is not promotional Chronos.Zx (talk) 06:53, 31 July 2025 (UTC)[reply]

Reporting DetectPromo false-positive

[edit]

Pulakeshin_II "boasts" in phrase "boasts, historic" is not promotional Chronos.Zx (talk) 07:56, 31 July 2025 (UTC)[reply]

Scripts++ Newsletter – Issue 27

[edit]

Reporting DetectPromo false-positive

[edit]

Pupil "dynamic" in phrase "dynamic, more than" is not promotional Chronos.Zx (talk) 13:30, 1 August 2025 (UTC)[reply]

Wikidata weekly summary #691

[edit]

Tech News: 2025-32

[edit]

MediaWiki message delivery 03:37, 5 August 2025 (UTC)[reply]