Wikipedia:Village pump (miscellaneous)

The miscellaneous section of the village pump is used to post messages that do not fit into any other category. Please post on the policy, technical, or proposals sections when appropriate, or at the help desk for assistance. For general knowledge questions, please use the reference desk.

Discussions are automatically archived after remaining inactive for a week.

« Archives, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81

An improved dashboard for the Content Translation tool

Hello Wikipedians,

The Language and Product Localization team has improved the Content Translation dashboard to create a consistent experience for all contributors using mobile and desktop devices. Below is a breakdown of important information about the improvement.

What are the improvements?
The improved translation dashboard allows all logged-in users of the tool to enjoy a consistent experience regardless of their type of device. With a harmonized experience, logged-in desktop users can now access the capabilities shown in the image below.

Does this improvement change the current accessibility of this tool in this Wikipedia?
The Content translation tool will remain in beta; therefore, only logged-in users who activated the tool from the beta features will continue to have access to the content translation tool. Also, if the tool is only available to a specific user group, it will remain that way.

When do we plan to implement this improvement?
We will implement it on your Wikipedia and others by 24th, March 2025.

What happens to the former dashboard after we implement the improvement?
You can still access it in the tool for some time. We will remove it from all Wikipedias by May 2025, as maintaining it will no longer be productive.

Where can I test this improvement and report any issues before it is implemented in this Wiki?
You can try the improved capabilities in the test wiki using this link: https://test.wikipedia.org/w/index.php?title=Special:ContentTranslation&campaign=contributionsmenu&to=es&filter-type=automatic&filter-id=previous-edits&active-list=suggestions&from=en#/

If you notice an issue related to the improved dashboard in the test wiki, please let us know in this thread and ping me, or report it in Phabricator, adding these tags: BUG REPORT and ContentTranslation.

Please ask us any questions regarding this improvement. Thank you!

On behalf of the Language and Product Localization team. UOzurumba (WMF) (talk) 17:56, 14 March 2025 (UTC)[reply]

Thanks for the update! —Ganesha811 (talk) 21:03, 18 March 2025 (UTC)[reply]

This isn’t related to the improvement, but about the content translation dashboard, do you know if there’s any way to get it to stop autofilling new paragraphs with the foreign-language text? Currently when I click “add paragraph” it automatically copies the French text, presumably as an alternative to machine translation, which I must then delete. It would be useful to be able to disable this, as en WP has disabled machine translation, and obviously having French in the final En article is not helpful. Mrfoogles (talk) 15:31, 26 March 2025 (UTC)[reply]

@Mrfoogles, there is a feedback button in Special:ContentTranslation. You might try using that.

Alternatively, perhaps some system could be set up to allow trusted users access to machine translation, which would help with (e.g.,) links. Send me an e-mail message if you'd like to know how to get machine translation. WhatamIdoing (talk) 03:27, 29 March 2025 (UTC)[reply]

I asked free ChatGPT "How, as a Wikipedia editor, would you respond to the person who wrote the original post in the discussion at https://en.wikipedia.org/wiki/Wikipedia:Village_pump_(miscellaneous)#I_boldly_put_LLM-generated_summary_suggestions_on_the_talk_pages_of_the_68_most_popular_articles_with_Technical_templates". After it replied, I asked it to reformat its reply as wikicode to be pasted here. Here it is, for what it's worth:

== Response to LLM-generated summary suggestions ==

As a fellow Wikipedia editor, I appreciate your initiative in exploring the use of Large Language Models (LLMs) to generate summary suggestions for the talk pages of high-traffic articles. Integrating advanced technologies like LLMs can potentially enhance our collaborative editing process. However, it's crucial to approach this integration thoughtfully, considering both the benefits and the challenges.

=== Key Considerations ===

1. Accuracy and Reliability: LLMs, while powerful, can produce content that appears factual but may contain inaccuracies or fabrications. Relying solely on LLM-generated summaries without thorough verification could lead to the dissemination of incorrect information. (Source)

2. Attribution and Transparency: When incorporating LLM-generated content, it's essential to clearly indicate its origin. This transparency allows other editors to critically assess the content's validity and ensures adherence to Wikipedia's commitment to verifiability. (Source)

3. Community Guidelines: The Wikipedia community is actively discussing the role of AI-generated content. Current guidelines advise against publishing content solely produced by LLMs without substantial human oversight and caution against using LLMs to create original content or references. (Source)

=== Recommendations ===

Collaborative Review: Encourage fellow editors to review and refine LLM-generated summaries. This collaborative approach leverages human expertise to validate and improve AI-generated content.
Ongoing Discussion: Engage in community forums to share experiences and gather feedback on using LLMs in the editing process. Collective insights can guide the development of best practices and policies.
Adherence to Policies: Stay updated with evolving guidelines regarding AI-generated content to ensure compliance and maintain the integrity of Wikipedia's information.

Your proactive efforts highlight the potential of integrating AI tools into our editing workflows. By proceeding with caution and fostering open dialogue, we can harness these technologies to enrich Wikipedia while upholding its standards of accuracy and reliability.

== Related Reading ==

Wikipedia is facing an existential crisis. Can Gen Z save it?

Largoplazo (talk) 22:04, 2 April 2025 (UTC)[reply]

More than 29,000 accounts compromised

WMF and User:AntiCompositeNumber has discovered and locked more than 29,000 compromised accounts. See m:Special:Log/WMFOffice. GZWDer (talk) 10:50, 27 March 2025 (UTC)[reply]

Additional information? How were they compromised? What kind of damage was done? — Rhododendrites ^talk \\ 14:08, 27 March 2025 (UTC)[reply]

Yes, please. A simple statement like "accounts using 2FA were not affected" would be helpful. ☆ Bri (talk) 19:05, 27 March 2025 (UTC)[reply]

Hi all. Yes, this is something we (WMF) are aware of and responding to. You can find more information on the Meta-wiki page m:Wikimedia Foundation/March 2025 discovery of account compromises; please add any questions on the talkpage there, keeping in mind that we cannot share some details for wp:beans/security/etc reasons. Also yes, confirming Bri is correct that accounts with 2fa were not affected. Thanks, Quiddity (WMF) (talk) 22:17, 27 March 2025 (UTC)[reply]

File:Liverpool FC.svg

Hi ,why this user publicated this logo (File:Liverpool FC.svg) for 2 articles, regarding the Wikipedia:Non-free content ,for logos used for 1 article but why the user publicated this logo for 2 articles?? (Google translator) AbchyZa22 (talk) 13:53, 27 March 2025 (UTC)[reply]

CAPTCHA

Somebody needs to shout this out loud.

MOST ASSISTIVE READERS CANNOT PASS THE NEW CAPTCHA TEST, BLOCKING ALL SUCH EDITORS FROM THEIR ACCOUNTS. PERSISTENT RANGEBLOCKS COVERING THINGS LIKE THE VODAFONE GATEWAY PREVENT EVEN IP ACCESS FOR MANY.

In protest, I am totally off-wiki from this moment on, unless and until this abuse of the disadvantaged ends. So no point in your replying to me personally. (Though I have passed a suggestion for a workable captcha to phabricator).

— Cheers, Steelpillow (Talk) 12:36, 28 March 2025 (UTC)[reply]

Thanks for the details of the new captcha test, the affected screen readers, and the phab ticket. ScottishFinnishRadish (talk) 12:45, 28 March 2025 (UTC)[reply]

Presumably the appropriate Phab link is phab:T6845#10686296, which is a new comment on a very old ticket. Anomie ⚔ 13:11, 28 March 2025 (UTC)[reply]

Per phab:T6845#10686370, sounds like they're about to turn it off. –Novem Linguae (talk) 13:28, 28 March 2025 (UTC)[reply]

Hey @ScottishFinnishRadish, I wanted to clarify that this particular change wasn't related with the tests. It was a separate decision, more like a one-off, that's also why we didn't check all accessibility consequences and rolled back that easily. SGrabarczuk (WMF) (talk) 23:26, 28 March 2025 (UTC)[reply]

I think that this should serve as a reminder to many people (including me) that accessibility is not a "nice-to-have" afterthought, but that systems are unusable to many people when it is not designed in from the start. Phil Bridger (talk) 14:16, 28 March 2025 (UTC)[reply]

Hey everyone, thanks for the comments above. We implemented CAPTCHA as a quick security measure: T390197 IPReputation: Support showing a CAPTCHA on Special:UserLogin and T379178 Support captcha as part of login flow (not just on "badlogin"). Given that it has an accessibility tradeoff, we did decide to roll it back. We will try to figure out a better way of securing logins keeping accessibility in mind. Thank you! SGrabarczuk (WMF) (talk) 18:16, 28 March 2025 (UTC)[reply]

Thank you for the prompt rollback. As something to work towards, may I suggest offering the alternative of 2FA by sending a confirmation code to the account's registered email address. Not perfect, as email is nowadays frowned on as less secure than phone nonsense, and some editors may not have registered, but better than what we have at the moment. Even more basic but a key part of the UX should be to add an accompanying caption stating that it is only a temporary measure; this would help inform those few who are locked out, meaning they will take it with far better grace than, say, I did. — Cheers, Steelpillow (Talk) 17:58, 30 March 2025 (UTC)[reply]

As it happens, we're working on that now and hope to have it deployed in some form soon, hopefully later this week T390437: Deploy Extension:EmailAuth. :)

I'm sorry for the issues we caused with deploying the CAPTCHA in this way, and thank you for speaking up about it. KHarlan (WMF) (talk) 18:54, 30 March 2025 (UTC)[reply]

Somebody needs to shout this out loud. Yeah, but it violates WP:SHOUT. Removing your bolding. Leaving the all-caps since I lack the energy to convert it. ―Mandruss ☎ IMO. 06:53, 2 April 2025 (UTC)[reply]

With due respect, WP:SHOUT says "seldom" not "never". I made that judgement call and I stand by it. The subsequent conversation shows I had a point. No violation there. FYI, I'm taking this off my watchlist now it's descended to trivia. — Cheers, Steelpillow (Talk) 12:54, 2 April 2025 (UTC)[reply]

You stand by it, yet you haven't restored your bolding. So the point of your comment was? Dude, if you feel that strongly about this, by all means shout to high heaven. I didn't mean to offend your sensitivities. ―Mandruss ☎ IMO. 14:23, 2 April 2025 (UTC)[reply]

Edit count of users blocked as LLMs increasing 10x per year

Please see this WT:LLM discussion.

Cramulator (talk) 01:37, 30 March 2025 (UTC)[reply]

Can I get credit for predicting this at Wikipedia:Eleventy-billion pool#2038 with my predictions for January 19?-Gadfium (talk) 02:53, 30 March 2025 (UTC)[reply]

Half credit. Cramulator (talk) 12:40, 3 April 2025 (UTC)[reply]

Welcome templates

I love all of them V1kor0to (talk) 05:12, 30 March 2025 (UTC)[reply]

Awesome job on the front page today

I really enjoyed the front page today. It's hard to get wordplay like that, and to have pretty much all the topics be so silly really made my day. Thanks to the folks who worked it, y'all knocked it outta the park. Wackogamer123 (talk) 02:42, 1 April 2025 (UTC)[reply]

It's just the DYK section on the front page that's April Fool's, right? Yeah, I see a couple funny ones in there :) –Novem Linguae (talk) 03:15, 1 April 2025 (UTC)[reply]

What if it all just ended…

Imagine if, one day, this encyclopedia were to go dark, if the Wikimedia project came to an end permanently. All the articles, all the discussions, the debates, the conflicts, the memories buried within its page all of it, gone forever.

How would you feel, after dedicating hours, months, even years of your life to it? After sharing so much knowledge, debating countless topics, forging connections with contributors from all over the world, what would remain, if Wikipedia suddenly vanished without a trace? Would it be sadness, nostalgia, or perhaps a quiet, lingering emptiness?

Riad Salih (talk) 11:37, 1 April 2025 (UTC)[reply]

I'm sure that, in the long term at least, those who only read would find something else to read, and the editors would find something else to do with their time. There would still be a need for some sort of online encyclopedia, but it would probably be very different. Phil Bridger (talk) 11:49, 1 April 2025 (UTC)[reply]

As part of the open source movement, if Wikipedia were to disappear, someone would surely fork it quickly using meta:Data dumps, and then folks would head over there. The work we've done is likely quite resilient to any kind of disappearance, attack, etc. –Novem Linguae (talk) 12:07, 1 April 2025 (UTC)[reply]

I'm sure there would be copies around from the regular data dumps. What would be hard to rebuild is the financial, infrastructure, and technical support supplied by the Foundation. There would also be the problem of making the public aware that a new clone was a legitimate successor to WP (depending on what might happen to the Foundation, the name "Wikipedia" might not be available for use). Donald Albury 15:37, 1 April 2025 (UTC)[reply]

I've worried about this recently, in light of the current abandonment of the rule of law in the US and the purge of everything and everyone contrary to the Administration's views. I hope that if they don't already replicate the site and its database to servers outside the US, say in Ireland or Denmark or Spain, that they're now arranging to do that. Largoplazo (talk) 13:31, 1 April 2025 (UTC)[reply]

The Foundation has several data centers around the world, although the ones outside the US are used for edge caching. Donald Albury 15:28, 1 April 2025 (UTC) Edited 15:41, 1 April 2025 (UTC)[reply]

The Federal government can enact legislation that in-effect makes life very difficult under the guise of combating piracy, pornography, anti-American propaganda. If the WMF is not already looking at moving the organization to Canada, Netherlands, etc.. they should be laying the foundation. The regime is cleaning house with a long wishlist of liberal crown jewels to dismantle. -- GreenC 04:02, 2 April 2025 (UTC)[reply]

I had actually thought about this two days ago when the site was having technical difficulties. Honest answer to your question? I would feel indifferent. As Phil Bridger said, people would find something else to read and something else do with their time. I know I would. I wonder how the wiki-addicts will cope though, but I'm sure they'll eventually get over it. Some1 (talk) 22:54, 1 April 2025 (UTC)[reply]

Wikipedia's intended purpose, to serve as an informational ready reference for most of the world, is beside the point? The damage would be measured only by its value as a leisure pursuit for the people who contribute to it and keep it running? Largoplazo (talk) 23:13, 1 April 2025 (UTC)[reply]

Some sadness, I love reading it as well as editing. The upside would be my new hobby which I'm rubbish at might improve. I've gone from putting out fires to creating them.

As long as YouTube survived? I've no idea how to fix things unless I can look it up on the toob. Knitsey (talk) 23:20, 1 April 2025 (UTC)[reply]

@Riad Salih For which reasons do you think this encyclopedia can cease to exist ?

If it does happens when I'm alive. I don't think I can predict how I will feel. Anatole-berthe (talk) 03:56, 2 April 2025 (UTC)[reply]

@Anatole-berthe just crossed my mind. But yeah, it can always happen. Just as an example, if Google somehow stops showing the encyclopedia in the top search results, it could seriously reduce the number of page views. It's a subtle way to kill something. Riad Salih (talk) 05:35, 2 April 2025 (UTC)[reply]

@Riad Salih I was thinking to the fact "Myspace" lost songs because of a faulty server migration.

There are many articles about this story.

An article among others.

It was wrote by "Matthew Robinson" published in "MARCH/18/2019" on the website of "CNN" : https://edition.cnn.com/2019/03/18/us/myspace-lost-12-years-music-uploads-apology-intl-scli/index.html

Your example about "Google search" seemed to me interesting.

In a hypothetical case like this that seem unlikely to me for the decade "2020s".

Maybe , a "Streisand effect" could save the encyclopedia instead of kill it. Anatole-berthe (talk) 06:21, 2 April 2025 (UTC)[reply]

The encyclopedia might become read-only (for what reason I won't hazard a guess), but I doubt it would disappear. If it disappears, read on. I poured my heart into software development for twenty years. A lot of it was really good software. I feel a sense of loss that none of it is likely to still be in use by mid-century, but such is life. I also miss music like what was being created in the 60s and 70s—a lot. I miss a lot of things. ―Mandruss ☎ IMO. 06:49, 2 April 2025 (UTC)[reply]

It's natural to miss and be personally affected with something one had dedicated time and efforts to. Many great projects had to become obsolete and be replaced with because they had over-grown and could no longer muster resources to go over the migration or just maintain themselves Nisingh.8 (talk) 14:57, 2 April 2025 (UTC)[reply]

No more charts in articles?

I think charts / datagraphics are some of the most useful in articles. Now it looks like some editors want top keep them out or remove them for being unreadable without clicking on them (don't know if this even applies on mobile where the images are displayed larger).

See Talk:Innovation#Image "too small to read"?.
I think datagraphics often contain very valuable information and should not be kept out by a few editors.
Prototyperspective (talk) 14:27, 2 April 2025 (UTC)[reply]

The discussion is about one chart in one article, not "No more charts in articles". Phil Bridger (talk) 15:09, 2 April 2025 (UTC)[reply]

What they say there extends to many and possibly most other datagraphics however. Prototyperspective (talk) 15:13, 2 April 2025 (UTC)[reply]

It's possible to have charts and datagraphics that work at the display size. The one at Innovation does not appear to do so however. CMD (talk) 15:25, 2 April 2025 (UTC)[reply]

Many or most charts need to be clicked on desktop or viewed on mobile (there it doesn't even need to be tapped but having to is not a problem if it's really informative). This is one of those many datagraphics in Wikipedia. If it was just one chart, then it would only show the adoption of one particular technology which wouldn't be (nearly as) useful or insightful. Prototyperspective (talk) 15:41, 2 April 2025 (UTC)[reply]

I boldly put LLM-generated summary suggestions on the talk pages of the 68 most popular articles with Technical templates

I have to start work very shortly, but I've been repeatedly urged to start a discussion here, so please pardon my brevity.

This morning, I posted summaries from the 68 most popular articles or their sections with the {{Technical}} template to their talk pages. So far I have about two positive, two neutral, and four negative comments. My major error was asking for fifth grade reading level summaries (because I had a vague recollection that was the target for the World Book Encyclopedia reading level) but I have since learned that STEM articles on Wikipedia are preferably written at the ninth grade reading level, when e.g. WP:ONEDOWN or the other conditions at WP:JARGON don't apply.

I did carefully read WP:LLM, which says, among many other pertinent things, that "LLMs can be used to copyedit or expand existing text and to generate ideas for new or existing articles." And generally, other pages such as Wikipedia:Using neural network language models on Wikipedia and this discussion seem even more positive on this use case. However, I agree that the summary should be sourced to the sources for the statements being summarized, and I don't know what I should do about that. I also read all of the 68 articles I generated summaries for and the summaries themselves. Anyway, here are the fifth grade and ninth grade level summaries, and I've released the source code into the public domain.

Please share your opinions! Cramulator (talk) 16:46, 2 April 2025 (UTC)[reply]

I think this was a reasonable idea. * Pppery * _{it has begun...} 16:47, 2 April 2025 (UTC)[reply]

Thank you. As is evident below, the vast majority of editors disagree, so I am immediately halting all non-userspace work on this project. I will post a short retraction of the suggestions to the 68 talk page sections in a day or two, barring any objections to doing that. In a week or two, I will write a postmortem report in my userspace in an attempt to summarize all of the comments received here, on the talk page sections, and on my user talk page, and link to it from the examples on WP:LLM and the other AI use guidance pages as a cautionary tale. Of course I will also post a link to that postmortem here.

I would like to thank everyone, especially those whose sensibilities I offended, for your frank and forthright criticism, and the words of support from the handful in the minority. There is so much I want to say about specific objections, some of which may never be able to be addressed adequately, some of which might be with varying degrees of difficulty, and some of which were probably based on misconceptions about my aspirations, the language I used when posting the 68 suggestions, the process used, what LLMs can and can not do, and other assumptions. That can wait. But no matter how much progress I can make improving the quality, style, appropriateness, presentation, and referencing of these suggestions, I will not be proposing to continue this project. Perhaps if I can show enough progress, someone else in coming years might.

I will say this in my defense: While I fully admit I have inadvertently caused disruption and anger far beyond what I expected, nevertheless, after just over half a day there are now five of the 68 articles which had been tagged as overly technical, all viewed by 500 to 2,500 readers per day, that have had their prose improved and their years-old cleanup tags removed. I will take some consolation that I may have spurred some help for those roughly half a million readers annually, from humans, not machines. Cramulator (talk) 06:42, 3 April 2025 (UTC)[reply]

In my opinion, bullshit-bot-generated text should be deleted from Wikipedia (articles, talk pages, everywhere) on sight. If a contributor can't communicate in their own words, they should leave such matters to people who can. AndyTheGrump (talk) 16:51, 2 April 2025 (UTC)[reply]

Please don't do this. It hurts to read these. I checked this ("A rainbow table is like a special cheat sheet that helps bad people guess passwords."), this and this ("If the minimum wage is already high, raising it more could make fewer people want to work"?). Fram (talk) 17:01, 2 April 2025 (UTC)[reply]

I'd like to comment on the Minimum wage#Welfare and labor market participation section summary. It is a poor summary of the final sentence, and a wholly inaccurate summary of the final paragraph, of the tagged section added by User:Mersenne56 in 2019, sourced to this book. I am unable to verify that the source actually supports the final sentence of that section. The plain language said to follow from the mathematics does not appear to me to follow at all. If prevailing wages are greater than those attainable by bargaining, an increase in the minimum wage should still increase labor force participation. I suggest the final paragraph is in error, and that the surprising statement was called out by the LLM because it is so counter-intuitive. I am not trying to make excuses for it; just understand why it happened. Cramulator (talk) 07:26, 3 April 2025 (UTC)[reply]

The statements I included are on p. 797-798 of the cited textbook. If the authors are wrong, so be it; however, I make no apologies for including material from a highly-cited textbook on labor economics. Mersenne56 (talk) 08:09, 3 April 2025 (UTC)[reply]

I am not sure that pages 797 and 798 support the final sentence of the section in question. Do you believe that they do? Cramulator (talk) 10:35, 3 April 2025 (UTC)[reply]

@Mersenne56: regarding this excerpt:

On the other hand, if w ≥ w*, any increase in the minimum wage entails a decline in labor market participation (because V_u decreases) and an increase in unemployment, which necessarily leads to a fall in employment.

What do you believe V_u represents? Cramulator (talk) 10:58, 3 April 2025 (UTC)[reply]

I fear imputing motive and awareness to that LLM, as in called out by the LLM because it is so counter-intuitive, might be an anthropomorphism that doesn't help us understand why it happened or generally use the LLM well. NebY (talk) 12:28, 3 April 2025 (UTC)[reply]

The LLM did not catch the problem, and neither did I. Fram did. Cramulator (talk) 12:41, 3 April 2025 (UTC)[reply]

Thanks for the clear demonstration that it's thoroughly inappropriate to generate summarires like that for Wikipedia. From your ninth-grade file:

to begin 5.56×45mm NATO#Cartridge dimensions: "The 5.56x45mm NATO cartridge has specific measurements and its case can hold 1.85 mL of volume."

to begin DisplayPort: "DisplayPort (DP) is a digital connection primarily used to link video sources, like computers, to display devices, such as monitors."

to begin Avicenna#Metaphysical doctrine: "Avicenna, an important Islamic philosopher, explored deep questions about reality, known as metaphysics."

Did you prompt it to write like a struggling ninth-grader? NebY (talk) 17:21, 2 April 2025 (UTC)[reply]

I looked through a few and found some striking issues. The whole of the summary on the Whore-Madonna complex I think would be unusable. It seems that the time spent making all these questions and discussing them could instead have been used to just read and summarize the articles. Sock-the-guy (talk) 19:21, 2 April 2025 (UTC)[reply]

As usual for LLM on Wikipedia, this appears to be creating more work than it is saving. Please discontinue. –Novem Linguae (talk) 21:10, 2 April 2025 (UTC)[reply]

Wikipedia content is required to be verifiable from reliable sources. If an editor pastes AI-generated content here, and they don't know where the LLM got the information from, they aren't in a position to vouch for it. If the LLM is asked to supply sources, there's still the matter of responsibility for vetting those sources and knowing whether the content the LLM attributes to those sources is actually in those sources and assessing whether the LLM has indulged in unpermitted synthesis. Ultimately, earnest Wikipedia editors out to be able to justify their contributions. The burden of doing so when the contributor didn't create the content seems pretty high. Largoplazo (talk) 21:49, 2 April 2025 (UTC)[reply]

There are ways to force it to use provided sources, and while there is still significant effort in double checking everything and doing some tidying it isn't too onerous. You can't just ask an LLM to do it, there's a process involved. Garbage in, garbage out. ScottishFinnishRadish (talk) 22:10, 2 April 2025 (UTC)[reply]

My opinion is the same as Andy's. There is zero value to any of this crap. Delete on sight. And make it absolute policy that LLM-generated text is not allowed on Wikipedia at all. And if someone spams that crap, they should be indeffed after one warning. oknazevad (talk) 01:38, 3 April 2025 (UTC)[reply]

Shit flow diagram is mostly LLM generated. I used chatgpt 4.5 with project mode and with the LLM only able to draw from specific sources as an experiment to see if there was any use cases. There were some things that required cleanup but it didn't do a bad job. It requires a fair amount of setup and guidance and obviously verifying everything, but it did a good job at accurately summarizing and citing the sources I provided. ScottishFinnishRadish (talk) 01:49, 3 April 2025 (UTC)[reply]

I don't give a damn. You managed to corral the bot, and did work needed to fix it up.Probably would have been easier just to write it from scratch. It's still starting with a fundamentally flawed source. I'd rather not have an article that have one that's unverifiable, unoriginal, and unneeded. oknazevad (talk) 15:03, 3 April 2025 (UTC)[reply]

These LLMs sound a lot like they know what they're on about... until they're on something you know about. And that's their biggest problem, even when used by well-informed editors. Less time spent writing content sounds great until you have to instead spend that time scrutinizing its output. XtraJovial (talk • contribs) 06:05, 3 April 2025 (UTC)[reply]

Although I'm not completely against use of AI to assist a well informed editor in writing, we should not be suggesting it's widespread use by the average editor with only a superficial understanding of a topic. I've used AI to help write professional material off-wiki in areas of my expertise. But like much technology, use by those without expertise risks serious errors. Wikipedia already fights an uphill battle with editors who have little understanding of topics they write about. AI can make writing appear impressive and give the writer a false sense of confidence, but if you look at professional reviews of its use it's scary. Sundayclose (talk) 17:05, 2 April 2025 (UTC)[reply]

I agree. I made similar points at Talk:Waste management where AI assumes that waste = trash. It doesn't. Velella ^{Velella Talk} 17:18, 2 April 2025 (UTC)[reply]

LLM-generated texts often violate Wikipedia guidelines. For example, both your 5th-grade and 9th-grade AI summaries of Existentialism#Facticity violate MOS:YOU. Phlsph7 (talk) 17:15, 2 April 2025 (UTC)[reply]

I'm going to hop in here, as it's related, to further publicize that I used ChatGPT 4.5 in project mode with sources uploaded directly for an experimental rewrite of Shit flow diagram. While converting all of the references to {{sfn}} I checked sources, verified content and rewrote some parts that had issues. ScottishFinnishRadish (talk) 17:24, 2 April 2025 (UTC)[reply]

These summaries are all really bad. LLMs are notoriously bad in particular at summarizing mathematical content. Tito Omburo (talk) 23:03, 2 April 2025 (UTC)[reply]

I'd have to suggest that rather than being 'notoriously bad' regarding maths, the bullshit bots are similarly bad at summarising anything else one has an actual understanding of. They merely give a superficial appearance of being competent in regards to stuff one doesn't have said expertise in. Observer bias... AndyTheGrump (talk) 23:18, 2 April 2025 (UTC)[reply]

That's fair and certainly keeping in my own experience. I didn't want to step out of my lane though. Tito Omburo (talk) 09:23, 3 April 2025 (UTC)[reply]

I agree that we shouldn't use these, but I thank @Cramulator for trying. AI is not going away, and it will only get better. Continuing to experiment with the technology is the best was to understand it and doing so on talk pages and asking for comments here was a responsible way to do that. RoySmith (talk) 11:23, 3 April 2025 (UTC)[reply]

True, but even then I'm not sure "better bullshit" will automatically translate into a net positive for the encyclopedia. Tito Omburo (talk) 11:32, 3 April 2025 (UTC)[reply]

We aren't trying to improve the quality of bullshit, we're putting the LTA vandals' ideas into a big wood chipper, and spreading the detritus on the side of the house, because we believe doing so can improve the mean quality of articles. Cramulator (talk) 12:44, 3 April 2025 (UTC)[reply]

I've tried to improve the prompt to source the suggestions. The output is at User:Cramulator/Summaries9a.json. Cramulator (talk) 14:13, 3 April 2025 (UTC)[reply]

I am really unsure what is being achieved here. The proposed summary for Waste management is almost identical to the existing first four paragraphs of the summary. What differences there are could be considered to be cosmetic. It is certainly true that the later paragraphs in the summary should probably be moived in shortened form to the body of the article as this summary has gradually becoming more bloated over the years, but the key paragaphs are so similar that I suspect that the LLM has simply read the original summary and thought "That will do nicely". I cannot see any value added benefit arising from this approach. Velella ^{Velella Talk} 14:36, 3 April 2025 (UTC)[reply]