Wikipedia:Village pump (policy)

"WP:VPP" redirects here. For proposals, see Wikipedia:Village pump (proposals).

 Policy Technical Proposals Idea lab WMF Miscellaneous 

• Table of contents
• First discussion
• End of page
• New post

The policy section of the village pump is used to discuss already proposed policies and guidelines and to discuss changes to existing policies and guidelines.

• If you want to propose something new that is not a policy or guideline, use Village pump (proposals).
• If you have a question about how to apply an existing policy or guideline, try one of the many Wikipedia:Noticeboards.
• If you want to ask what the policy is on something, try the Help desk or the Teahouse.
• This is not the place to resolve disputes over how a policy should be implemented. Please see Wikipedia:Dispute resolution for how to proceed in such cases.
• If you want to propose a new or amended speedy deletion criterion, use Wikipedia talk:Criteria for speedy deletion.

Please see this FAQ page for a list of frequently rejected or ignored proposals. Discussions are automatically archived after remaining inactive for two weeks.

« Archives, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195

Don’t move to draft perfunctory!

You should not move articles to draft space just for them being new and unfinished!

I created an article and it was moved to draft almost immediately, Draft:Raksila Artificial Ice Rink Pakkalan kenttä. The reason given was, that there were no sources. OK, fine, I understand. So I add sources. It’s done in an hour. Now, to have the article un-draftified, it will likely take months, or so it at least says in the information box provided.

Wouldn’t it be better to just contact me and ask me to add sources to the very new article I had written? (If you feel you cannot do it yourself.)

I think draft ought to be used only for articles which we have given some time to become better, with reasonable suggestions to the writer, but the article still hasn’t become better. Bandy långe (talk) 23:07, 15 November 2022 (UTC)

@Bandy långe, articles done directly in main space should be reader-ready from the beginning. Start articles either in a user sandbox or in Draft space. You can move them to main space yourself when they are ready. Review is optional for autoconfirmed users like yourself. StarryGrandma (talk) 19:30, 16 November 2022 (UTC)

@StarryGrandma, so? Please comment on the topic in stead of giving advice not asked for. Wikipedia does not work the way you seem to think, articles are not created ready from the start. The whole point is that this encyclopedia is a collaborative effort where people write things together. Bandy långe (talk) 22:52, 16 November 2022 (UTC)

@Bandy långe, you seem to have missed the part where StarryGrandma says you don't have to wait for months for it to be undraftified, you can just move it to mainspace yourself. And I'm curious why you would publish an article to mainspace before adding the sources you used to write it? JoelleJay (talk) 02:04, 17 November 2022 (UTC)

She did not say that. She said that if I had started the article in draft or in a sandbox, I could then move it to main space myself, not if someone else moved it to draft. Bandy långe (talk) 18:23, 17 November 2022 (UTC)

It doesn't matter how it got to draftspace, you can always move it back to mainspace as an autoconfirmed user. JoelleJay (talk) 19:58, 17 November 2022 (UTC)

If I'm reading the timestamps right, the article was in mainspace for seven hours before being draftified. This is well above the minimum guidance at WP:NPP. Also, if you object to the draftification, you should just move it back to mainspace using the more menu, then clicking move. Hope this helps. –Novem Linguae (talk) 06:57, 17 November 2022 (UTC)

No, it says it should stay in draft space until it has been reviewed. As far as I understand the information, you are not even supposed to approve the article while it is there (but I did that in a way anyway, by adding the sources; I suppose I should appologise for that). It is therefor generally considered better, I suppose, to have it in draft spce limbo for months than to just remind the writer to add references to it. Do you mean I am not obliged to wait for a review? Won’t it just be moved back to draft if I overrule the person moving it to draft by moving it back? Bandy långe (talk) 18:23, 17 November 2022 (UTC)

No, it says it should stay in draft space until it has been reviewed. As far as I understand the information, you are not even supposed to approve the article while it is there (but I did that in a way anyway, by adding the sources; I suppose I should appologise for that). Where are you finding this very incorrect information? None of the policies or guidelines related to draftspace say any of that. JoelleJay (talk) 20:01, 17 November 2022 (UTC)

This is a good demonstration of Wikipedia's unreliability. First, the OP treats this site as a blog, posting something or other without any backing. Because, the sources will-appear (?) soon (?). Then, the blogger expects others to come and validate the "contribution", doing the hard work. However, as anyone that ever contributed facts in Wikipedia knows, the wikitext and its sourcing are intimately linked, and both the substance and the presentation of an article depend on the context and the content of the sources supporting it. And that is even before applying conceptually higher-level policies such as NPOV and impartiality.

I can't see how one can offer a coherent article without simultaneously adding a minimum of basic sources. Off to draftspace, come back when the article has something to say. 71.247.146.98 (talk) 13:25, 17 November 2022 (UTC)

You got it wrong. Maybe it’sbecause you are a casual visitor to Wikipedia, not having a login, so you don’t know how this works. It is the person who sends an article to be reviewed in draft space, who thinks someone else should take care of it. He thinks the person who started the article should not continue working on it, but that it should be reviewed by others within some months. You should read about what Wikipedia draft is. Bandy långe (talk) 18:23, 17 November 2022 (UTC)

@Bandy långe, you are conflating "review" in the New Page Patrol sense (as is linked in the draftification notice) with "review" in the WP:AfC sense. NPP reviews all articles created by non-autopatrolled editors, but a review is not required to remain in or move a draft to mainspace. AfC is a completely different process for submitting articles into mainspace and is not necessary for autoconfirmed users. You can bypass AfC completely by moving the article yourself.

As for He thinks the person who started the article should not continue working on it, I have no idea where you got this idea. The draftification notice explicitly says you can continue working on the draft: Your article is now a draft where you can improve it undisturbed for a while. JoelleJay (talk) 20:22, 17 November 2022 (UTC)

§ Don’t move to draft perfunctory!

You should not move articles to draft space just for them being new and unfinished!

It seems you object to your mainspace post being drafted in the first place. Your draftspace-related lecture is a newer thing. You claim to know something that should be published in Wikipedia. We want to know why. Don't reply to me or anybody; just add reliable references. There, situation resolved. 65.88.88.59 (talk) 22:25, 17 November 2022 (UTC)

@Bandy långe: Wikipedia articles in article space are often a work in progress. But to exist in article space they need to meet the "Is an article of this topic allowed to exist in mainspace?" criteria which is mostly/usually WP:Notability. And usually this means supplying 2 (maybe one) GNG type sources. It will inevitably get reviewed regarding this by New Page Patrol. IMO it is good practice and a reasonable expectation that new articles in mainspace (at least within an hour) include GNG type sourcing to establish wp:notability and that they reside somewhere else as a draft until they have that GNG sourcing. But if it meets those criteria, IMO it is OK to be in mainspace regardless of the amount of work needed in other areas. And when it meets those criteria, you can move it yourself, you don't need to wait. Sincerely, North8000 (talk) 19:03, 17 November 2022 (UTC)

Really? The information about the draft space and the need for draft review suggests otherwise, I think. Bandy långe (talk) 19:47, 17 November 2022 (UTC)

@Bandy långe What information about draft space and draft review are you reading/looking at that makes you think otherwise? Perhaps there's some policy page or guideline that needs to be clarified. ~ ONUnicorn^{(Talk|Contribs)}problem solving 20:01, 17 November 2022 (UTC)

Not sure even which part of my post they are talking about. If it's the "you can move it" part, I think that boilerplate text on one of the draftify templates falsely implies otherwise; maybe that is where Bandy långe's impression came from. North8000 (talk) 20:55, 17 November 2022 (UTC)

Editors that use this draftification script (including me) are given default language to use on the article creator's user talk page. It ends with "When you feel the article meets Wikipedia's general notability guideline and thus is ready for mainspace, please click on the "Submit your draft for review!" button at the top of the page."Firefangledfeathers (talk / contribs) 21:13, 17 November 2022 (UTC)

The OP has identified a common problem. Editors who have made 10 edits and whose accounts are more than a few days old have the ability to create articles directly in the mainspace, and also the ability to move articles out of the Draft: namespace. However, almost none of them know this.

I suspect that the more active reviewers don't actually want this to be well known. If they have to personally approve everything, then the mainspace will always meet their standards of being "reader ready". If (almost) anyone can move pages, and they know it, then the mainspace will inevitably be sullied by all of these WP:IMPERFECT articles. WhatamIdoing (talk) 20:57, 18 November 2022 (UTC)

The draftification script is deceptive/wrong. It says that when you're ready "click on the "Submit your draft for review!" button at the top of the page"" thus implying that that is THE (=only) next step. And most editors will assume that such is authoritative or based on some rule. BTW I think that you you made some pretty incorrect and negative assumptions about active reviewers. NPP reviewers mostly want to just get the review done, and "perfection" isn't the standard. I think that AFC reviews are a lot tougher only because the the system coerces the reviewers to be overly cautious and thus tough. A AFC approval is implicitly a stamp of approval of everything about the article. A NPP approval is implicitly less, something along the lines of "an article of this topic is allowed to exist in article space" North8000 (talk) 21:16, 18 November 2022 (UTC)

I think that AFC reviews are a lot tougher only because the the system coerces the reviewers to be overly cautious and thus tough. Might not even be this. AFC is similar to NPP except AFC doesn't do WP:BEFORE. AFC requires all sources to be evaluated to already be in the article. This makes it de facto slightly tougher than NPP. The rest is very similar. –Novem Linguae (talk) 21:55, 18 November 2022 (UTC)

I think that what you said is structurally true and how it should be, but not true in practice. I think that in practice, an AFC approval is defacto sort of the reviewer signing off that the overall article has no significant problems, which is a broader and tougher standard than NPP. While a NPP'er may tag an article for other quality issues, the reasons for failing an article are much narrower. Another reason for this is that failure at NPP is a more "severe" act (AFD or the reviewer moving the article out of article space etc.) than an AFC reviewer just saying that it needs more work and re-submittal. North8000 (talk) 22:17, 18 November 2022 (UTC)

I agree with North about the incentive structure, but both review systems (NPP and AFC) suffer from this. Nobody wants to be the "bad reviewer" who "approved" a "bad article". I've seen articles declined because:

• the sources weren't in English.
• the refs weren't formatted using ref tags.
• a navbox is "promotional" (it looked like the list of names at the top of Template:Disney).
• the newly appointed CEO of Disney isn't notable.
• a 1300-word-long news article entirely about the subject doesn't show notability (it's in-depth, secondary, independent, and reliable, but still not enough for that reviewer).
• thousand-year-old National Treasures of Korea haven't demonstrated notability.
• BLPs should have sources at the end of every sentence.

I've also seen reviewers decline autobiographies, mostly of young people who think that being one of the millions of aspiring professional musicians or athletes should make them eligible for an article; I list only obvious errors here, not the everyday, run-of-the-mill decisions to decline articles.

To give you an illustration of the misaligned incentives here, consider Draft:Richard Winkler (Producer) which Greenman declined today. It is a decision that is both correct and incorrect. The subject "has won seven Tony Awards and five Olivier Awards", which pretty much guarantees notability. No Broadway producer going to win that many Tonys and Oliviers and not get written up in at least his hometown newspaper (that'd be the Detroit Free Press, if anyone wants to search), and a dozen "well-known and significant awards" easily clears WP:ANYBIO. AFC is supposed to accept anything that is likely to survive AFD, and this will. But the sources that are presently cited in the article are unimpressive, and precisely because his name is on everything, it's going to take more than a couple of seconds with a search engine to find sources that do more than just mention his name and his role in a play. Worse, there appears to be a television producer with the same name, so nearly all of the sources contain no more than a single sentence about this Winkler, some are about the other Winkler, and you're trying to find the tiny fraction that actually help you write a whole article. So you decline it, because the cited sources are unimpressive, and it will take some effort, probably by someone who knows which sources to focus on if you want to know more about a theater person, to improve the sourcing, but you probably shouldn't because it is about a notable subject, except if you accept it, then someone might yell at you about the weak sourcing, but if you don't, you're declining a BLP that met ANYBIO twelve times over, over a problem that can be solved through normal editing, but...

You can see the bind we're putting the reviewers in. There is no action this reviewer can take that everyone will agree is correct. We also don't have a culture of deference towards the reviewers making decisions on borderline cases, the way we do with admins at WP:AE. The only thing we can say with reasonable certainty is that there is more scrutiny on accepting than on declining, so you're slightly less likely to get yelled at for declining a weak article on a notable subject than for accepting it. WhatamIdoing (talk) 03:29, 29 November 2022 (UTC)

All of what @WhatamIdoing said is exactly why it's easier for the original author to include the sources that the draft or article was based on, from the start. Surely, @Bandy långe, you didn't write the article out of thin air, and you have ths sources readily at hand, correct? If you have edited here long enough to create articles directly in article space, it looks like you are expected to know the requirements. Just my thoughts. David10244 (talk) 07:04, 1 December 2022 (UTC)

Technically, avoiding speedy deletion per {{db-person}} requires that there be a Wikipedia:Credible claim of significance in the Wikipedia article itself, not just in the cited sources. One might also have sources that support ANYBIO notability (e.g., Variety (magazine) saying only "Winkler has won seven Tony Awards" in passing), but the reviewer might be looking for GNG-style sources. Merely including the sources from the start doesn't solve the problem. WhatamIdoing (talk) 20:30, 1 December 2022 (UTC)

I consider myself reasonably intelligent, but I was submitting articles for years before I discovered that I could polish my work in draft and then move it to mainspace. It was the wording of the draft template that threw me. Just saying!! Downsize43 (talk) 07:54, 1 December 2022 (UTC)

This is a common gap in knowledge, and we are not really incentivized to tell editors the truth about it. (You can also polish your work in a userspace sandbox, which is the choice that most experienced editors seem to make.) WhatamIdoing (talk) 20:31, 1 December 2022 (UTC)

Using user page subpages for article development is covered in Wikipedia:User pages#Terminology and page locations, but new users have a lot of other P&G pages they should look at, as well, so they may miss it. Donald Albury 21:10, 1 December 2022 (UTC)

Galactica and RS

Platform 9¾ at King's Cross Station

Meta's Galactica seems to be able to rapidly generate WP pages, although at present they're going to be relatively easy to identify as fake. Presumably they're going to get better in the future.

My question is really about the fake references it might generate and how we are going to better protect ourselves against fake content, fake notability and fake RS. For me, one of the great weaknesses of AfD discussions has always been the possibility of printed RS which exist on a dusty library shelf. If we have AI that can generate plausible looking book references, isn't it going to be an increasing challenge to identify completely fraudulent pages? JMWt (talk) 10:02, 17 November 2022 (UTC)

Well how are fake but plausible-seeming references generated by organic intelligence dealt with now? I wouldn't overwork myself trying to find out. There is no formal validation of citations for accuracy or relevance in Wikipedia, and there is no other metric that will help the answer. It is left to the community to haphazardly and erratically certify references, at least outside of vanity projects like so-called "good" or "featured" articles. If anything the pre-seed of native AI present now (the relationship of relevant Wikidata properties with article verification policies/guidelines) when applied is likely to make things worse, as there is no context validation of Wikidata data to begin with. 65.88.88.68 (talk) 16:22, 17 November 2022 (UTC)

And on the other hand, there's Assigning Numbers. RAN1 (talk) 20:40, 18 November 2022 (UTC)

@JMWt, is your concern that the machine-learning system will write a book, that book will get published somewhere/by someone, and the book will be cited in Wikipedia? WhatamIdoing (talk) 21:00, 18 November 2022 (UTC)

@WhatamIdoing I think it is more plausible that some machine learning system generates references that look like very old books that would take a lot of effort to check. I don't think it needs to get to the stage of actually publishing anything to be a problem. JMWt (talk) 08:25, 19 November 2022 (UTC)

The question posed was not answered: why are machines producing articles with inappropriate references a bigger concern than humans doing so? Some of the latter may be doing so now, undetected. And does it matter what kind of entity publishes misinformation? In any case compiling an encyclopedia is a mechanical process, there is nothing creative about it. Non-human machines will be able to replicate it easily. 69.203.140.37 (talk) 14:36, 19 November 2022 (UTC)

@JMWt, if the machine-generated pseudo-book isn't published, then how would a Wikipedia editor have access to its contents?

69.203, at some point, quantity becomes its own quality. A human who types all day without stopping (e.g., to interview anyone or to double-check facts) can usually only produce a volume of text equal to about one book a week. A room full of computers could produce a book a minute without stopping. WhatamIdoing (talk) 22:11, 20 November 2022 (UTC)

Wasn't there a recent convoluted discussion/RFC about mass article production/publishing articles at scale? 64.18.11.71 (talk) 01:56, 21 November 2022 (UTC)

@WhatamIdoing Galactica can generate convincing looking references. An editor could just machine-generate a whole WP page including the refs. All completely bogus. Maybe I'm missing some detail that you are asking me? JMWt (talk) 11:21, 21 November 2022 (UTC)

@JMWt, Are you concerned about ghost references ("nothing remotely resembling the content appears in this real source") and hoax citations ("There is no Platform 9¾ at King's Cross Station, so your content can't be supported by a source located there")? WhatamIdoing (talk) 22:52, 21 November 2022 (UTC)

@WhatamIdoing Galactica was, I think, creating completely bogus references. If they are recent, we can probably identify them with a search of the ISDN or DOI. If they are old, that's going to be nearly impossible. It might also be copying real references and claiming they contain facts that they don't. Both are a problem, no? JMWt (talk) 07:12, 22 November 2022 (UTC)

Given how much research has been done (for five or ten years now) on matching newspaper articles to Wikipedia statements, I wonder why they would bother creating a hoax citation. Just to see if the software could mimic the style, with plausible content? But they could plug in the other system, and get real citations to real sources. There'd be some limits on accuracy ("A traffic problem of nightmarish proportions occurred when Joe Film stopped to sign an autograph and a clamoring crowd of thousands formed around the star. Police were called charged the actor with Pranks without a Permit in the third degree": a simple software program could find "Joe Film" and "traffic", but might not be able to figure out whether it should be written to suggest guilt, innocence, or even whether it's worth including.) WhatamIdoing (talk) 17:08, 22 November 2022 (UTC)

@JMWt Apropos of this, Stack Overflow have explicitly - for the moment - banned the use of a similar tool (ChatGPT). They specify that specific one and not the general class of tools, though it might just be that this is the only one usable at scale at the moment - Galactica lasted about a week so presumably isn't much of a problem any more! Andrew Gray (talk) 14:10, 5 December 2022 (UTC)

Child safe searches

should wikipedia searches be child safe 23.115.40.162 (talk) 00:21, 23 November 2022 (UTC)

Generally, this will be not the case because of the policy WP:REDACTION and I don't expect you'll see any momentum to change that. If this is important to you, the best idea is probably a client-side solution like web filtering software. Jahaza (talk) 03:51, 23 November 2022 (UTC)

The very first problem with something like this defining what is meant be "child safe" in some objective way, and the second problem is getting consensus on a single definition given that everybody has different ideas about what children should and shouldn't be able to see. One common suggestion is basing it on something like articles in Category:Sex. However, that would cover articles that most people find innocuous such as ZW sex-determination system, Golden calf, List of female Nobel laureates and Wildflower. Restricting it to the top-level of the category but not subcategories would block Female and Sex segregation but not Sexual intercourse or Paedophilia. Thryduulf (talk) 15:10, 23 November 2022 (UTC)

@Thryduulf is correct that there is no universal definition of what "child safe" means. For example, the general consensus in the United States seems to be that violence is OK, but nudity is not. I would argue exactly the opposite. I'm sure you have your own idea of what's acceptable and what's not.

I like the way IMDB does this. Rather than assign ratings, reviewers describe specific things which are depicted in the films. "'Balls' is said once in a vulgar context." "Several positive characters are killed or mortally wounded on-screen during the movie, which may be disturbing for the audience.", "we see their backsides ... quite a bit". I think there's a great opportunity for a third-party rating service along these lines. Build a database and invite people to describe potentially objectionable aspects of articles. Expose these ratings though a publicly accessible API. Now, somebody could build a search tool which sits on top of Mediawiki's own search API and uses your ratings API to provide filtering according to criteria you select. "Don't show any articles which contain pictures of human nudity". "Don't show any articles which describe violence". "Don't show any articles which contain any of this list of words".

I don't believe content filtering is something wikipedia should be involved in. But our licensing and API availability make it possible for people to do their own. It's not a trivial undertaking, but there's nothing that's fundamentally difficult about it. A proof-of-concept would be a reasonable semester project for an undergraduate software engineering course. The hardest part is getting enough quality crowd-sourced data, but that would be true no matter who implements it. -- RoySmith (talk) 15:45, 23 November 2022 (UTC)

And What Wikipedia is not#Wikipedia is not censored is policy. I will note that section of the policy was originally entitiled "Wikipedia is not censored for the protection of minors", so the policy that we do not censor the contents of Wikipedia to protect children goes back to the very early days of Wikipedia, and would require a major shift in one of the fundamental policies of the project. Donald Albury 15:21, 23 November 2022 (UTC)

Any sort of search where things were censored would have to be external, as we are WP:NOTCENSORED. Lee Vilenski ^{(talk • contribs)} 15:30, 23 November 2022 (UTC)

If you want a version of Wikipedia that is safe(er) for kids, try Kiddle. They take our freely licensed content, censor and rewrite it so that it is more kid-friendly, and republish it under a free license. ~ ONUnicorn^{(Talk|Contribs)}problem solving 15:40, 23 November 2022 (UTC)

Fifteen years ago, when I was running the library/computer center/network at a small school, I added a similar "kid-friendly" WP clone site to the school's computers. Good to know that something similar is still around. Donald Albury 16:58, 23 November 2022 (UTC)

Tracked in Phabricator
Task T306246

Perhaps the OP is referring to the new and "improved" search bar in Vector-2022. Go to https://en.wikipedia.org/wiki/?useskin=vector-2022, and try searching for "analgesic". But type very slowly, letting the page update after each letter. Readers (whether children or adults) should not have NSFW content WP:GRATUITOUSly forced on them like that. Search for the name of a body part, click on the article, and you should expect to see a picture of that body part. But not any page with a common prefix. Suffusion of Yellow (talk) 22:33, 23 November 2022 (UTC)

Yeah, we should be filtering search suggestions, as I think Google and others do, to avoid this problem. I haven't looked but I bet there's already a phab ticket and it's probably years old. Levivich (talk) 02:21, 25 November 2022 (UTC)

The challenge with filtering is that you'd have to start classifying articles as ineligible for inclusion in search autosuggest, and I'm not sure if there's already an easy classification to re-use. You could specify certain categories, but those are always subject to change and it would be easy for anyone to troll autosuggest by adding/removing the relevant categories from their article of interest. The major search engines do this with a combination of algorithmic analysis and human-curated lists, and I foresee both practical issues and editor drama from trying something like that here. Taking the "analgesic" example above, I see that analgesic does appear immediately below the potentially NSFW result. If people are truly searching for "analgesic" on a large scale, then a better ranking implementation would go a long way to solving this issue without getting into filtering and classification. Orange Suede Sofa (talk) 02:40, 25 November 2022 (UTC)

I agree, ranking is probably better than filtering, but I don't think it will be difficult to come up with a list of naughty words and phrases to exclude from autosuggest. And also, it wouldn't be hard to have a "safe search" toggle like Google has, either, so people can filter or not filter results. And we could turn safe search on for school IP ranges. Levivich (talk) 03:17, 25 November 2022 (UTC)

Let's please not go down the road of automatic censorship for schools. When a school kid wonders if their penis is normal and comes to us looking for pictures to compare against, we should be showing them pictures of penises. Because if we don't, they'll go elsewhere to find the information they seek.

Google operates under a different set of constraints than we do. Google makes money by selling advertising displayed as part of their search results. When Walmart says, "We won't spend any of our huge advertising budget here if our ads show up next to pictures of penises", Google listens. And they build tools like "safe search" to placate advertisers like Walmart. That's not an issue here. -- RoySmith (talk) 04:05, 25 November 2022 (UTC)

To be clear, I was talking about the images that will soon appear in the search the suggestions on desktop, and already appear on mobile. I don't think links to any article should be hidden from the suggestions. But showing File:Wiki-analsex.png to everyone who searches for a subject starting with "ana"? Suffusion of Yellow (talk) 04:31, 25 November 2022 (UTC)

Hmmm, from reading phab:T306246, it looks like we might be able to take care of that ourselves, through MediaWiki:Pageimages-denylist. Suffusion of Yellow (talk) 04:45, 25 November 2022 (UTC)

Strongly disagree with the idea of censoring encyclopeic content or supporting any third-party efforts to do so. Wiki articles are not movies, and any effort to assign movie-style labels or ratings to articles would inevitably be misused to deny access to articles like sexual intercourse in public libraries and other institutions. –dlthewave ☎ 13:59, 26 November 2022 (UTC)

Standard for harmful content on user pages

There's been some debate regarding what constitutes harmful content in userspace and what's considered disruptive or inflammatory. Personally, I've leaned toward not allowing any sort of strong political expression, seeing it as inherently disruptive. I'd like to suggest a compromise:

Users may not advocate or endorse the violation of any rights described in the Universal Declaration of Human Rights.

This declaration of human rights was written in 1948 in response to the Holocaust, and it's the most widely agreed upon list of human rights. Advocating the violation of human rights would be a very obvious way to determine whether a user has crossed a line, and it aligns with the current policies against Nazism, sexism, pro-slavery, etc. To respond to what I expect would be the primary concern: no, it's not a precise standard. There would still be discussions about what is and isn't acceptable. But it would be significantly more precise than the current standard of gut feeling and WP:IDONTLIKEIT, giving a clear foundation for these discussions that currently doesn't exist. Thebiguglyalien (talk) 21:13, 25 November 2022 (UTC)

pinging WaltCip because you've posted about this problem a few times. Thebiguglyalien (talk) 21:14, 25 November 2022 (UTC)

My personal line for this sort of thing is having/advocating content on user pages that is negative toward groups of people. Though I could also see extending it toward political content. But the Universal Declaration does seem fine as a baseline. Silverseren^C 21:16, 25 November 2022 (UTC)

Unless one feels UDHR is another worthless feelgood statement, good only for decorum, legal and political grandstanding and for making gullible people believing there is/will be some sort of "progress". Let's pretend that most of the signatories have not once violated any of the enumerated rights since the document was signed. One may realistically consider it a failure, just like its sponsoring institution. It is interesting that it was proposed only a few years after civilized, well-educated and democratically elected people used nuclear weapons (twice) on civilian targets. This is not an anti-American statement, or even a political one, it is factual. That fact (that universal human rights can easily be violated by well-meaning people) never meaningfully entered the discourse around that document. The fact being ignored, keeps reappearing, over and over. 104.247.55.106 (talk) 21:55, 25 November 2022 (UTC)

One of said rights is 'freedom to hold opinions without interference'. Does this include the right to hold opinions incompatible with the Declaration? I think we may have encountered a paradox... AndyTheGrump (talk) 22:06, 25 November 2022 (UTC)

The paradox exists only because the document exists. Otherwise any opinion is just that, an opinion. There was a quote by Brecht about calling an overflowing river "violent", when the banks that constrain it define the overflow. 104.247.55.106 (talk) 22:33, 25 November 2022 (UTC)

This brings us back to a fundamental debate about human rights: are they divinely granted and endowed upon us by a Creator, or are they conferred by a document or a rule of law? Are they revocable at any time or are they innate and irrevocable, even when unrecognized and violated? Elizium23 (talk) 09:52, 27 November 2022 (UTC)

Unfortunately, most people haven't read the Universal Declaration of Human Rights, much less ponder its contents or its message, so I doubt this would be a very effective rule. I'm with AndyTheGrump on this one. But moreover, I think any potential RfC focused on userboxes with political or divisive content needs to focus on being an all-or-nothing affair, specifically to prevent posses from running MfD and determining what is and isn't allowable. I don't see there being much compromise available beyond that. 🌈WaltCip-(talk) 00:42, 26 November 2022 (UTC)

I bet i could get my userpage deleted just by selective quoting of the Articles. fiveby(zero) 01:03, 26 November 2022 (UTC)

That likely won't fly, considering that the Islamic world rejected the UDHR and drafted the Cairo Declaration on Human Rights in Islam in its place to remove referenced to religious freedom. 2603:7080:8F02:2B11:D13C:AE3B:86FB:FB29 (talk) 15:16, 26 November 2022 (UTC)

IP vandalism guideline

I want to make IP vandalism a guideline. I have already made the page. SpyridisioAnnis Discussion 15:39, 26 November 2022 (UTC)

This (assuming we're talking about Wikipedia:IP vandalism) is redundant to the vandalism policy. firefly ( t · c ) 16:12, 26 November 2022 (UTC)

That's because there should be multiple vandalism guidelines about different ways of vandalizing Wikipedia SpyridisioAnnis Discussion 04:40, 27 November 2022 (UTC)

One such guide is enough, unless any type of vandalism is too complex to handle that covering it in the one guide would make the guide confusing or otherwise unreadable. Animal lover |666| 06:20, 28 November 2022 (UTC)

RFC Related to Reliable Sourcing

I'm requesting input at Talk:Jason_David_Frank#RFC:_Cause_of_Death, specifically on a dispute related to Wikipedia's reliable-sourcing policy; one user has suggested a few implications of the policy that I don't think are supported by text or practice. TMZ has reported that, per anonymous "law-enforcement sources," Jason David Frank died by suicide. The aforementioned user has said that this report cannot be included on the JDF page. There are three aspects of this dispute, all related to WP:RS.

1. Does WP require an official statement from family or law enforcement? (According to the user: "we 100% . . . need an official statement from his management team or law enforcement.")
2. If an authority itself relies on anonymous sources, can that authority be cited? (According to the user: "Doesn't matter if it's a death, sports transaction, or whatever per that policy, no citations based on anonymous sources.")
3. Is TMZ an unreliable source? (Note: WP:TMZ, an essay, has some thoughts on that subject; it recommends explicitly attributing information to TMZ.)

Given how differently he and I read existing policy, I think third-party perspectives would be helpful.--50.86.94.196 (talk) 08:09, 27 November 2022 (UTC)

Briefly,

1. No, however we do require that WP:RS of some kind to have covered a cause of death for it to be included.
2. Depends, does the source have a reputation for high-quality fact-checking and responsible reporting? If so we can probably trust they've done the legwork, but situationally editorial discretion always applies (sources split, only reported in one source especially if the claim is surprising/important, current events/developing news etc.)
3. Currently classed as WP:MREL which more or less rules out using it to support controversial WP:BLP (also applies to recently deceased) information. You can always ask for additional input at WP:RSN if you believe that the source has been misclassified, or that one particular piece should be viewed differently from the publication as a whole.

For future reference the help desk and teahouse will usually provide more prompt responses to inquiries. If you are looking for input into a dispute that involves you and one other editor you can also follow the process outlined at WP:3O. Since this particular case already has an associated RFC there really isn't any additional need to request input on matters of content; hope this helps, 74.73.224.126 (talk) 19:27, 6 December 2022 (UTC)

Hi! Appreciate the tip. I was following the suggestion on WP:RFC to publicize the RFC on the Village Pump ("To get more input, you may publicize the RfC by posting a notice at one or more of the following locations: One of the Village Pump forums, such as those for policy issues, proposals, or miscellaneous"). I think the publicizing did its job! Got quite a bit of input. Thanks though!--50.86.94.196 (talk) 23:41, 6 December 2022 (UTC)

No problem, in general the intent is for WP:APPNOTEs publicizing discussions to be brief; often just using standard templates such as {{rfc notice}} or {{please see}} to avoid forking the discussion. And of course all ongoing rfcs are automatically listed at WP:RFC/A. That said it looks like you did get plenty of additional input; all's well that ends well I suppose. 74.73.224.126 (talk) 01:22, 7 December 2022 (UTC)

"Deleted image removed" - how long is long enough?

While perusing punctuation errors, I found that Strongwoman has a not-uncommon feature in Wikipedia articles, a block of hidden text saying up front, "Deleted image removed". In 2009. Can we please have a bot scour all these ancient deleted images and purge them from the Wikitext altogether? BD2412 T 23:05, 27 November 2022 (UTC)

Not sure there is any policy/guideline/etc requiring these to be made, retained, or removed. They are still actively being added (e.g. Special:Diff/1124102240) by User:ImageRemovalBot - operated by User:Carnildo who may have more background on this. There are about 20,000 articles like that. — xaosflux ^Talk 23:34, 27 November 2022 (UTC)

I have no objection to the addition when an image in the article is deleted. I just think that there's a shelf life beyond which they are no longer useful, and become nothing more than pollution of the Wikitext. I would say that after a few years (perhaps two or three, even), the notice has outlived any utility it may have initially had. BD2412 T 23:48, 27 November 2022 (UTC)

Is there any actual need to remove these comments? If not, they should probably only be removed as part of an other edit, and certainly not by a bot. Animal lover |666| 06:15, 28 November 2022 (UTC)

The need to remove these comments is that useless strings of hidden wikitext make it more difficult to find and edit the wikitext that needs editing. From my own experience, I fix a lot of punctuation errors, and would prefer not to be bothered finding such errors in hidden text, where they crop up for some reason. BD2412 T 17:45, 28 November 2022 (UTC)

If this is going to happen, one thing to consider going forward would be to ask the bot to add the date to the comment (e.g. Deleted image removed 20221128: ...) to make it easier on future updates. — xaosflux ^Talk 17:49, 28 November 2022 (UTC)

Maybe that information should be added to the talk page instead of the wikitext. ~ ONUnicorn^{(Talk|Contribs)}problem solving 17:51, 28 November 2022 (UTC)

I can see the value in noting where in the article an image has been removed from, as it makes it easy to add a new one or restore the old one if it is undeleted for some reason. However, the value in that decreases as the article develops without the image there - time is a proxy for that but I don't expect it's a very consistent one. There have been 54 revisions to Strongwoman since the image was removed, looking at some other articles the bot edited that day (20 July 2009) there have been thousands (so many it doesn't tell me the number) of revisions to Manchester Arena but only 9 to Type 518 radar. I think it would be rarely problematic to remove the text when it has been in place for the greater of 2 years and 50 edits, or where the article has since become a GA or FA (although how likely the text is to still be there in them I have no idea).

Timestamps are good idea, and posting on the talk page as well as the article also seems sensible. Thryduulf (talk) 21:43, 28 November 2022 (UTC)

@ONUnicorn I suggested in the wikitext, as that would make a future bot cleanup job much easier (as it would have the data there - as opposed to having to try to scrape the history to see how long it was there). Now another possibility would be to have that bot not do this at all anymore, and instead just post it to the article talk page. Not sure if that would be even less useful though? — xaosflux ^Talk 21:50, 28 November 2022 (UTC)

I can see the use of such a hidden note for a short period. An editor currently working on the page notices that an image has disappeared, clicks to edit to see if there is something wrong in the wikitext, and sees a note explaining that the image has been deleted, so they don't have to chase after what happened to it. I would say that if such notes are going into the wikitext, they should be there for two years at most (no matter how many edits have been made to the page, if no one has bothered after a deleted image for multiple years, the note will be of no further value). BD2412 T 17:14, 29 November 2022 (UTC)

I agree with the general point here. That aside, there's a fairly strong community aversion to bot edits that have no effect on the appearance of a page as seen by readers. Essentially the upside (cleaner wikitext) is perceived as outweighed by the downside (watchlist flooding, additional page history rummage). So there would need to be a rather strong case for why this is somehow qualitatively different; even linter-fixing bots engaged in future-proofing have proven controversial. 74.73.224.126 (talk) 19:08, 6 December 2022 (UTC)

The original reason for replacing the images with something was to avoid breaking table layouts. I don't know if that's still a problem or not. The specific replacement with a comment was as a result of a discussion during OrphanBot's approval back in 2005; when I split OrphanBot's tasks between ImageTaggingBot and ImageRemovalBot, ImageRemovalBot inherited the behavior. --Carnildo (talk) 19:30, 29 November 2022 (UTC)

Global deleters addition at Global rights policy

Turns out m:Global deleters Is a thing... Was just boldly added to Wikipedia:Global rights policy. This was after a 2 hour discussion at the Bureaucrats' noticeboard. If you are interested: Wikipedia:Bureaucrats' noticeboard#Request for permission to run Synchbot Terasail^[✉️] 21:12, 3 December 2022 (UTC)

Should global deleters be permitted to delete local pages when fulfilling m:Synchbot requests?

This would be made effective with the section currently {{proposed}} at Wikipedia:Global rights policy#Global deleters. This follows the above-mentioned thread at WP:BN.

 Note: There is (currently) only one global deleter (who previously carried out this function with local admin privileges). –xeno^talk 23:21, 3 December 2022 (UTC)

• I would prefer wording this in a slightly different way ("should the English Wikipedia allow Synchbot to delete pages", incorporating the way Synchbot works by reference), but otherwise support this. * Pppery * _{it has begun...} 23:37, 3 December 2022 (UTC)

  Synchbot is only a concept, an ideal… –xeno^talk 23:42, 3 December 2022 (UTC)

• Yes — pre-existing global user right (2014) designed for this use case — TheresNoTime (talk • they/them) 23:40, 3 December 2022 (UTC)
• Reluctant yes. Yes as far as the WP:GRP goes, this works and is fairly uncontroversial. Reluctant in that I don't like this process and think it needs to be overhauled, but that needs to happen upstream, and upstream from there at the lack of developers - and I don't think we should break this fairly rare process here in the meantime. — xaosflux ^Talk 23:56, 3 December 2022 (UTC)
  • The very fact that it's rare - about 30 deletions a year - means that "breaking" it is of minimal harm; and I don't see how it's been broken at commons or arwiki anyway, which have both for many years required Synchbot to place a speedy tag instead of just deleting. —Cryptic 00:56, 4 December 2022 (UTC)
• Yes per xaosflux. --Rschen7754 00:01, 4 December 2022 (UTC)
• I'd be less uncomfortable with this if m:Synchbot made it explicit that it won't delete pages that have been moved into userspace. I trust Pathoschild, as a former enwiki admin, already checks for that; I have no way to know if whoever else starts up a new Synchbot-like service in some nebulous future after Pathoschild goes inactive will think to. —Cryptic 00:56, 4 December 2022 (UTC)
• Yes to this and the CSS/JS section below. No preference on wording; I'm not really fussed about how we choose to define the rules for one trusted person and their ~30 uncontroversial deletions a year. Giraffer ^{(talk·contribs)} 09:35, 4 December 2022 (UTC)
• If there's also a bot approval for it Besides the question as to whether the bot should be allowed to use its global permission here, there's also compliance with WP:BOTPOL and Wikipedia:Global rights policy#Global bots to be considered here. Anomie⚔ 14:42, 4 December 2022 (UTC)

  Anomie: Should this be done even though it's not a bot in the traditional sense? Although on Meta it is marketed as a bot, in practice it is a user running a semi-automated script on their own account. –xeno^talk 15:20, 4 December 2022 (UTC)

  @Anomie there is no bot here, this is just a script that is letting a person script some of their (rather infrequent) actions. The only person that ever does this doesn't do it unattended either. — xaosflux ^Talk 15:54, 4 December 2022 (UTC)

  @Xeno and Xaosflux: How odd that there's a bot account with the name, and this RFC is talking about allowing the bot to operate, and the linked topic at WP:BN is titled "Request for permission to run Synchbot", but the bot doesn't actually perform the task. If a human continues doing it in a semi-automated manner, then ok. If it becomes an actual bot at some point, then said bot would need a BRFA. Anomie⚔ 23:35, 4 December 2022 (UTC)

  There's no bot account with that name. Synchbot is a bot in the general sense of 'a tool or script which performs actions', but not in the enwiki policy sense of 'an account with a bot flag' (similar to other tools like Pywikibot). So 'Synchbot' is both the custom bot library (which performs the edits/deletions) and the abstract service (which users submit requests for), and I use that library through my regular wiki account to fulfill those requests. The naming dates back to 2008 when I guess that usage was more common. —Pathoschild (talk) 01:03, 5 December 2022 (UTC)

  @Anomie this entire discussion so far has nothing at all to do with the account Special:CentralAuth/Synchbot, which is a doppelganger that only exists to avoid having someone register it and cause confusion with the script of the same name that we are talking about. — xaosflux ^Talk 01:14, 5 December 2022 (UTC)

  Sure. Except where the BN thread specifically said "Request for permission to run Synchbot", and some above said similar things. Anomie⚔ 01:57, 5 December 2022 (UTC)

  @Anomie I think we can agree the names are confusing! "Synchbot" is an action automation tool, it runs under the credential of a logged in user; the only person that ever uses it (pathoschild) has some global permissions that mostly let it work on all WMF wiki's. That person was until very recently an admin here, who resigned in the face of upcoming inactivity requirements. They would like to keep using it here, using their global permissions. They asked on BN, even though that really isn't the right venue - but they were following up on their resignation -- which has led us to here. — xaosflux ^Talk 02:20, 5 December 2022 (UTC)

  I'm glad I'm not the only one that got ridiculously turned around about this situation. Primefac (talk) 13:09, 8 December 2022 (UTC)

• This should be allowed. These three points define enough of a scope for the rights use. Terasail^[✉️] 15:30, 4 December 2022 (UTC)
• Yes, and strikes me as something that should be relatively uncontroversial given the minimal risks compared to the benefit that the service provides. -- Ajraddatz (talk) 19:51, 4 December 2022 (UTC)
• Support per NOTBURO. HouseBlaster^talk 17:33, 6 December 2022 (UTC)
• Yes this process has been running uncontroversially for years and the global group was created specifically for it. I'm not too bothered over the exact wording; Pathoschild can be trusted to act within community expectations. 74.73.224.126 (talk) 18:47, 6 December 2022 (UTC)
• Yes, but considering the global right has existed for years already and had the approval of meta, I'm not even entirely sure that we as the enwiki community have the authority to say no. casualdejekyll 22:49, 7 December 2022 (UTC)

Interface editing

Another task Synchbot does is make edits to user CSS/JS pages. This is even rarer than deletions, and was historically allowed by virtue of his global interface editor access, but, by a strict reading of Wikipedia:Global rights policy#Global interface editors is now prohibited since Pathoschild has previously had the administrator or interface administrator right removed at the English Wikipedia. I suggest changing the quoted section, and while I'm at it the substantively identical wording at Wikipedia:Global rights policy#Global rollbackers, to previously had the administrator or interface administrator right for cause removed at the English Wikipedia * Pppery * _{it has begun...} 00:28, 4 December 2022 (UTC)

Agree: this makes sense to me. –xeno^talk 00:37, 4 December 2022 (UTC)

Shouldn't that be "removed for cause", or is this an ENGVAR issue? —Cryptic 01:13, 4 December 2022 (UTC)

Should be "removed for cause", that was a typo. * Pppery * _{it has begun...} 01:29, 4 December 2022 (UTC)

"Inactivity" is a cause. — xaosflux ^Talk 01:15, 4 December 2022 (UTC)

But I'm fairly open to supporting an update to that that would get around voluntary resignations, or even just procedural inactivity removals alone. — xaosflux ^Talk 01:18, 4 December 2022 (UTC)

I was thinking "for cause" in the sense that Wikipedia:Former administrators/reason/for cause (and the mainspace redirect for cause) uses it, not "for any cause at all", which would be obviously useless. * Pppery * _{it has begun...} 01:29, 4 December 2022 (UTC)

I agree with Xaosflux that I would prefer not to use "for cause" as a blanket term that is assumed to exclude procedural removal of user rights, if there is consensus for a change. I think for clarity it would better to list the specific exceptions desired. isaacl (talk) 02:15, 4 December 2022 (UTC)

"for any reason other than inactivity" (the wording used at WP:EFH) would seem to work here. If there are other reasons for admin removal then they can be spelled out too. Thryduulf (talk) 14:50, 4 December 2022 (UTC)

You would also need to explicitly exclude resigning (since that's what Pathoschild technically did). I'm still not convinced of the need to spell this out any further than my initial proposal, though. * Pppery * _{it has begun...} 15:13, 4 December 2022 (UTC)

• OK so GIE's can edit here if they were never admins, assuming they follow all other types of policies and don't do things that make people mad.... We certainly don't want GIE's operating when the editor was already desysoped for cause related to negative behaviors here. Inactivity isn't really the same, since if they just never bothered to become an admin here it wouldn't apply. Since we're touching the GRP\GIE section how about a bigger change
• FROM:
  • Global interface editors can use their rights by default, provided they have not previously had the administrator or interface administrator right removed at the English Wikipedia. If removal has previously occurred, they must request and be granted interface administrator and/or administrator access locally by an English Wikipedia Bureaucrat. Furthermore, any English Wikipedia bureaucrat can ask a global interface editor to stop using their global privilege if what they deem to be misuse occurs, and the global interface editor must comply with such a request. Such a decision by a bureaucrat can be appealed to the wider community. Failure to comply may result in a block. Interface editors are reminded that some interface pages, such as the watchlist, require discussion prior to use.
• TO:
  • Global interface editors canmay use their rights by default, provided they have not previously had the administrator or interface administrator right removed at the English Wikipediainvoluntarily, other than procedurally for inactivity. If such a removal has previously occurred, they must request and be granted interface administrator and/or administrator access locally by an English Wikipedia Bureaucratprior to using this access. Furthermore, any English Wikipedia bureaucrat can ask a global interface editor to stop using their global privilege if what they deem to be misuse occurs, and the global interface editor must comply with such a request. Such a decision by a bureaucrat can be appealed to the wider community. Failure to comply may result in a block. Interface editors are reminded that some interface pages, such as the watchlist, require discussion prior to use. Interface editors must comply with local policies, such as the protection policy.

I think that clears up this issue, and takes care of some housekeeping. — xaosflux ^Talk 21:45, 4 December 2022 (UTC)

Under this proposed rewrite, users who voluntarily resign their advanced user rights, would still be allowed to use their global rights, even if the resignation was voluntarily given while the user was under a cloud. This doesn't sit well with me, as this would essentially create a loophole whereby editors who lose EnWiki community trust could nevertheless still take interface admin actions on EnWiki by simply resigning.

Something along the lines of Global interface editors canmay use their rights by default, provided they have not previously voluntarily resigned the administrator or interface administrator right under a cloud or had the administrator or interface administrator right removed at the English Wikipediainvoluntarily, other than procedurally for inactivity. would solve this problem. — Red-tailed hawk _(nest) 21:13, 5 December 2022 (UTC)

In such situations the bueraucrat who pushes the button to remove admin or Iadmin access can invoke the any English Wikipedia bureaucrat can ask a global interface editor to stop using their global privilege if what they deem to be misuse occurs clause. Since clouds are currently determined by crats at the time of resysop that would be more consistent with the way adminship works.

To be clear, I have no objection to any rewording that makes it clear this doesn't apply to Pathochild (or Mr. Stradivarius, who technically violated the aforementioned clause with this series of edits in October 2022 after losing IADMIN for lack of use in August 2022), but I don't see this line of working as necessary. * Pppery * _{it has begun...} 21:28, 5 December 2022 (UTC)

I think this is covered by WP:IAR and WP:NOTBURO; we don't need to change policy due to a single exception. BilledMammal (talk) 21:59, 4 December 2022 (UTC)

If a user had previously held a right, and would get it back on request, then they should be allowed per IAR to operate a bot as if they had gotten the right back. If a request would need to be discussed first, then the discussion must precede getting the right to operate such a bot. Of course, this doesn't override any other rules or restrictions on operating bots, such as approval of the task itself. Animal lover |666| 06:05, 5 December 2022 (UTC)

As of January 2023, Pathoschild won't be able to get his admin bit back on request. That's the entire point. * Pppery * _{it has begun...} 21:28, 5 December 2022 (UTC)

I agree with the proposed changes above, we note that consensus at AN (or some other venue) can restrict a user from exercising the privileges granted here. HouseBlaster^talk 17:33, 6 December 2022 (UTC)

Justapedia: the far-right historical revisionist Wikipedia FORK

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.

---

• File:Justapedia creator Justme edit to Nazism article.jpg
  • caption1 = Edit to Nazism
• File:Justapedia creator Justme edit to Template Nazism let wing.jpg
  • caption2 =Edit to Template:Nazism left wing
• File:Justapedia creator Justme edit to Template Nazism let wing editnotice.jpg
• caption3 =Edit to Template:Nazism left wing editnotice

Yesterday I posted a message at ANI and it was suggested this was a more appropriate place to alert the community about Justapedia. Basically it is a proposed online encyclopedia being marketed as "the neutral and objective encyclopedia that Wikipedia should have been". Yesterday the website was taken down after the post at ANI.

The creators of Justapedia appear to be active Wikipedia editors who, unhappy with the way this community works, have download the entirety of English Wikipedia with the intention of marketing it as their own. They have even plagiarised the majority of Wikipedia's policies and guidelines. Since the post at ANI yesterday they seem to have taken their website down for now.

Where their project starts to diverge from Wikipedia appears to be American politics and administration. In terms of content, the overwhelming majority remains what contributors to English Wikipedia have made. But, they are removing critical commentary of conservative US political figures and engaging in some far-right historical revisionism, for example claiming Nazism is a left wing ideology that is comparable to contemporary US Democratic Party ideologies, on the right are some screenshots of some Justapedia diffs that were taken before visibility was restricted. In terms of project administration, they seem unhappy with community consensus and instead intend to retain complete control through a Politburo-like "Board of Representatives", while all editorial decisions will be enforced by a number political commissar-like appointees to the "Editorial Board".

• File:Justapedia creator Justme user TP comment 1.jpg
• File:Justapedia creator Justme user TP comment 2.jpg
• File:Justapedia creator Justme user TP comment 3.jpg

The creators have even started their own foundation and of course an associated fundraising campaign, a promotional video has been uploaded to YouTube and it is being marketed on social media. I recommend going to YouTube and searching for "Justapedia". Apparently the Earthwave Society is sponsoring them for now [1]. It appears the same person founded both organisations, I assume they are Justapedia editor User:Justme, to the right is a screenshot of the since deleted user page comment that led me to believe this.

How does the community feel about active contributors here attempting to distort the ideology that led to the Holocaust for their own shallow political ends? And how does the community feel about these same Wikipedians attempting to profit from the years of hard work of the community? Justapidiot (talk) 07:21, 4 December 2022 (UTC)

I'm not sure how 'the community' feels about it actually matters much, since the right to fork content has been built into the project from the start. That's how a Creative Commons license works. As long as content is properly attributed, it can be copied. And this isn't new - people have been creating Wikipedia forks for years. They rarely last long. AndyTheGrump (talk) 07:33, 4 December 2022 (UTC)

Is it being properly attributed? Silverseren^C 07:38, 4 December 2022 (UTC)

Hard to tell, with the website down. Thought even if it isn't, actually doing anything about it isn't easy, since copyright for edits remains with individual contributors, and the WMF can't act on their behalf: see Wikipedia:Mirrors and forks. AndyTheGrump (talk) 07:44, 4 December 2022 (UTC)

I have blocked this account since the username references a controversy and insults people who have forked Wikipedia, which is legitimate. I have no idea who this person is, but it seems clear to me that this is probably an illegitimate use of a sock account in an attempt to evade scrutiny. Cullen328 (talk) 07:57, 4 December 2022 (UTC)

@Cullen328 I have filed a DR on c:Commons:Deletion_requests/Files_uploaded_by_Justapidiot for these screenshots. Lemonaka (talk) 10:53, 4 December 2022 (UTC)

The block was fine but I dont think there were trying to evade scrutiny as they are clearing receiving scrutiny and their username was clearly made in order to attract scrutiny. Qwv (talk) 12:29, 4 December 2022 (UTC)

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.

Controver... articles

I am review some parts of wiki due to other issues that I am working and reporting on and I note several different title format issues that in most cases do not follow and particular logic or are not linked to the correct article or category and other issues. I have had a look and can find no guidance or policy, and as an IP user I cannot just rename them to something more suitable, and in many cases I cannot edit, to add to the appropriate interested group, container cat etc. the solution I feel would be update policy to cover these article, search provided below and then add to the various spaces as required. [[2]]

Policy would need to say something like avoid the use of the words controversy, controversial etc in the title (except specific music, book, etc titles), these article must be linked to or placed in the See also of the main article and that the title must begin with the subject focus, so, as an example Controversies of the Hong Kong Police Force, would become Hong Kong Police criticisms or whatever is agreed by the wider team. The goal is to place these potentially leading article in front of all interested parties to promote clean-up, better referencing, merging, redirects and avoid invalid unwanted duplications etc 2404:4408:638C:5E00:B099:F55F:A773:FF11 (talk) 10:15, 5 December 2022 (UTC)

Editing under an open proxy

It is prohibited to edit while using an open proxy. I was trying to edit under a proxy server to detour censorship in my country.

It it totally understandable to block editing under a proxy but I think it is to strict.

For example, editing a page of in one's own namespace may not be a problem. In addition, a logged-in user (not an anonymous IP user) with considerable experience (either in time or edit counts) may be not so vulnerable.

Shouldn't the rules about the proxies be relaxed a little bit? Regpath (talk) 00:10, 8 December 2022 (UTC)

@Regpath you can see the rules about editing with proxies here: Wikipedia:Open proxies. There is no block on reading articles via proxy. — xaosflux ^Talk 00:38, 8 December 2022 (UTC)

@Xaosflux Thanks. I already read that article and know that reading is not limited. I wanted to discuss 'editing' policy under proxy. Regpath (talk) 00:42, 8 December 2022 (UTC)

An experienced logged-in user can be granted the ability to edit via anonymizing proxies by asking the checkuser team for an exemption from IP blocks—see Wikipedia:IP block exemption for more information. Generally, this privilege is only granted to editors who demonstrate a need for it, e.g. because their country censors Wikipedia. Mz7 (talk) 03:50, 8 December 2022 (UTC)

Given that iOS devices have the option of turning on the "private relay" which is an open proxy I think that this is going to start affecting more people over time. Gusfriend (talk) 07:16, 8 December 2022 (UTC)

Wikipedia response to chatbot-generated content

• ChatGPT has gone viral and is causing quite a storm on the Web. Why should we care? Because, rather than just providing links to an article for you to read about what you are interested in, it writes an essay about it on the spot. It is so sophisticated that, if you ask it to write an article on any subject, even in the style of Wikipedia, it will! That can then be copied and pasted into Wikipedia -- not necessarily a problem if it is carefully fact checked and edited first (ChatGPT can't tell fact from fiction), but, an editor may post it without doing so. It can even be used to answer questions about Wikipedia as if it is a Wikipedian (see the first link provided directly below).    — The Transhumanist   06:52, 9 December 2022 (UTC)
  1. Experiment: See it being used to generate answers to hypothetical questions on Wikipedia:Teahouse
  2. ChatGPT's Writing Capabilities Stun, but Humans Are Still Essential (for Now) - CNET
  3. Five Chats to Help You Understand ChatGPT - The Atlantic
  4. The ChatGPT chatbot is blowing people away with its writing skills. An expert explains why it's so impressive
  5. ChatGPT: Why Everyone's Obsessed With This Mind-Blowing AI Chatbot - CNET
  6. ChatGPT | Discover AI use cases
  7. 'Scary' ChatGPT could render Google obsolete in two years
  8. ChatGPT, an AI chatbot, has gone viral. Some say it’s better than Google; others worry it's problematic.
  9. What ChatGPT AI is and how you can use it - The Washington Post
  10. What is ChatGPT and why does it matter? Here's what you need to know | ZDNET

• Based on how rapidly chatbots have improved over time, it will become more and more difficult to tell if an article was written by a computer or not. The sheer volume at which computer programs could create new accounts and produce Wikipedia content, and the inevitable growing number of human editors copying and pasting chatbot output into Wikipedia, will at some point make it impossible for Wikipedia's human volunteers to keep up with that traffic and apply quality control to the material in a reasonable time frame -- the backlog of unchecked material will simply get longer and longer. The only recourse will be for computer programs to do it -- either computer programs to process articles to filter out or correct any crap, or training the chatbots themselves not to produce crap in the first place. Rather than build computer algorithms to detect computer-written articles and passages, it would be more productive for them to do style checks, fact checks, and citation checks, along with appropriate corrections or removals. While Wikpedia-friendly AI could come from within Wikipedia, it may be faster to bring influence to bear upon the developers of the chatbots being used to generate Wikipedia content, and upon the chatbots themselves. Wikipedia already has a chair at the table, because Wikipedia comprises a significant component of chatbot corpi, and so, their developers should be inclined to listen to the Wikipedia community's concerns -- either directly, or indirectly through news coverage. The Wikipedia community should make its voice heard on the matter of chatbots writing Wikipedia material according to Wikipedia's style and behavior guidelines. For example, verifiability still applies, and so when chatbots are asked by their users to "write an article in the style of Wikipedia" the chatbots should comply according to Wikipedia's policies, including those on verifiability and providing reliable sources. Not doing so should be met with the filing of bug reports, feedback, and commentary. And, as chatbots learn as they go, Wikipedians who use them can ask them to follow Wikipedia guidelines, and we can urge our fellow editors to request this of chatbots as well.    — The Transhumanist   06:52, 9 December 2022 (UTC)
  • Chatbots should be following Wikipedia's lead for all of their output. At this time, most chatbot answers and essays are not referenced with reliable sources. And they should be, for the same reason that Wikipedia articles should be. That's something that can be requested of chatbots directly, through queries, and of developers, through their contact channels and social media. I hope this suggestion helps.    — The Transhumanist   06:52, 9 December 2022 (UTC)
• The simple answer is that our existing policies ought to already cover this (mostly.) Sourcing is still required for anything that is challenged or likely to be challenged, which prevents people from just blindly dumping AI generated text into Wikipedia; and an AI may violate copyright depending on how it was trained (and whether it was overtrained.) There are also unsettled copyright concerns related to AI training sets, so I would generally think that, ideally, editors shouldn't be dumping AI generated text into our articles even after performing due diligence to make sure it's not a copyvio and finding proper sources. But since those concerns are unsettled and speculative, I also don't think it's worth worrying about too much right now. The key point is that we should emphasize our sourcing requirements and be more diligent for clear-cut copyvios, which we already have systems in place to handle, since it is likely that these tools will result in people adding lots of unsourced and possibly-copyright-violating text. (I do wish our RFCs on mass article creation had reached a stronger agreement on sourcing requirements for new articles, which would deter excessive copy-pastes of AI generated text - perhaps that is something we might want to revisit in the near future, if we start seeing significant amounts of new unsourced articles created using what is plainly AI-generated text.) --Aquillion (talk) 07:55, 9 December 2022 (UTC)
  • You mean, don't prepare in advance for a potential increase in volume, just wait until it hits? At that time, will merely adjusting policies stem the tide? It's in the slow trickle phase now, but that could potentially become a torrential flood very rapidly, just as ChatGPT's user base grew to over a million in 5 days. My main concern above was about a potential volume of AI-generated content that went beyond the scale of what the editor community could manually process. You didn't address that contingency. What could the community do to prepare for it, just in case it does happen? What are the available options?    — The Transhumanist   11:28, 9 December 2022 (UTC)
    • I don't think there's much we reasonably can do to prepare, at least not without serious risk of causing other problems; AI-generated text won't be drastically different than other sorts of text, aside from the risk of being uncited or a copyvio (which we have existing processes in place to handle.) It's worth raising awareness of the issue so editors can spot the signs of someone using large amounts of it, but I think our best bet if we're going to "prepare" is to focus on the systems we already have, which is unlikely to do any harm either way, or perhaps to codify slightly more strict sourcing requirements in the way I described (which I think is a good thing anyway, but would at least serve to slow down the worst sorts of misuses of AI generated text.) Ultimately the most serious problems are if editors start adding large amounts of text that violates copyright or which are uncited and likely to be challenged, but we have existing procedures for those, we just need to prepare for the possibility that we may need to become a bit more aggressive about enforcing them. Wikipedia is in a slightly better position than some other websites facing AI-generated-text problems, because our sourcing requirements will at least make it fairly obvious if someone tries to dump large amounts of AI-generated text onto the wiki without making any effort to verify it. --Aquillion (talk) 12:47, 9 December 2022 (UTC)

      I suppose we could take the Stack Exchange approach and just say flatly "no, this isn't allowed" - in their case it is explicitly a temporary measure until we have a better understanding of the issues. I think in general our policies/community norms would come down hard on anyone trying to get a language model to generate articles (hard to see why that would be OK and machine-translation isn't), but maybe an explicit statement would be a way to go. Andrew Gray (talk) 18:32, 9 December 2022 (UTC)

You seem to be soewhat panicking over a scenario which isn't really supported by any evidence. While I see some Teahouse responses, could you give us one or two examples of " It is so sophisticated that, if you ask it to write an article on any subject, even in the style of Wikipedia, it will! " articles? The teahouse examples give the impression that, if it ever becomes a problem, some edit filters can easily spot these. You would in any case need "someone" to post this "potential volume of AI-generated content that went beyond the scale of what the editor community could manually process" you predict. This seems rather unlikely, at least on enwiki. Fram (talk) 11:45, 9 December 2022 (UTC)

• Just try it. If your eyes don't pop out, I'll be surprised. Right now, during its "research preview", it is free. Keep in mind that it draws heavily on Wikipedia, which is included in its corpus, so, for this test run, it would be best to choose a person or subject that is not yet covered in this encyclopedia, and ask ChatGPT to write about that.    — The Transhumanist   14:41, 9 December 2022 (UTC)
  • Pinging @Fram, JPxG, EpicPupper, and 0xDeadbeef:    — The Transhumanist   14:50, 9 December 2022 (UTC)
    • I'm not giving out my phone number to some random website, thanks. Why it isn't sufficient that they have my email which was then verified is not clear... Fram (talk) 14:55, 9 December 2022 (UTC)

      LOL I had the same exact response. My phone number? F no. Levivich (talk) 16:55, 9 December 2022 (UTC)

      I went to try this out, and it asked me for my phone number. I thought about making one up like 0118 999 881 99 9119 725 3, but figured it would probably use it for two factor authentication, so that's no good. Ritchie333 ^{(talk) (cont)} 18:31, 9 December 2022 (UTC)

• @Fram, Levivich, and Ritchie333:     I wasn't that bright. I gave it a random phone number. It rejected it as a land line. Then I gave it another, and it rejected that as a VOIP number. Finally, I gave it a random mobile phone number, and it sent some complete stranger the verification code. Oops.    — The Transhumanist   01:32, 11 December 2022 (UTC)

• • Thanks for the ping. I'd imagine the biggest problem would be people using the AI to create hoaxes. Like the Zhemao hoaxes but with less effort. 0xDeadbeef→∞ (talk to me) 15:13, 9 December 2022 (UTC)

  So. I tried it yesterday. I'm not sure how heavily it draws on Wikipedia's corpus for its knowledge.

  • First, I asked it to tell me about Hammerton Killick. I know there is a Wikipedia article about Hammerton Killick, because I wrote 90% of it. It did not know who Hammerton Killick was, and informed me that it does not have access to the internet, or to Wikipedia.
  • Next, I asked it to write me an article in the style of Wikipedia. I did not specify a subject. It wrote about Athens. The result was ok. Heavily focused on the ancient city and on art and architecture. Short. Kind of read like an encyclopedia article.
  • Next, I asked it to write me an article about alcoholism in the style of Wikipedia. The result was very interesting. I did not think it read like a Wikipedia article, it was more like a brochure that would be distributed in a doctor's office or something. I asked it what about that essay it thought was like Wikipedia, and it said what it wrote was
    • neutral
    • factual
    • organized
  • Next, for fun, I asked it if it could write a recipe. It proceeded to give me a recipe for chocolate chip cookies. It looked like it should work. I e-mailed it to myself, and today I made them, not expecting much. I was pleasantly surprised. They were delicious. The only problems with what it wrote was that it did not have me cook them long enough (it said to bake for 8-10 minutes, and it took closer to 13 minutes for them to be done), and it drastically underestimated how many cookies the recipe should make (it said I'd get 2 dozen cookies, and I ended up with 5 dozen). I was shocked that it actually was edible.
  • I asked it to write a legal motion asking the court for an in-person hearing. I did not give it any other details. For not having any other details, the result was not bad. Westlaw has started offering a service that I think might draw on this type of technology, it helps you write pleadings.
  • Last I asked it to write a 100 word short story about a mouse, genera: fantasy. The result was decent. If I came up with it on my own I wouldn't be ashamed to enter it into a contest like the ones NYC Midnight runs.

  I was more impressed with the recipe and the short story than the Wikipedia style articles. I can see some use for it in, say, copyediting as JPxG did below; or asking it for suggestions on language rephrase if you are trying to reach a word limit. I think it could have its uses. But I do think the Wikipedia community should be looking to craft policies and guidelines around what is and is not acceptable use of such tools. ~ ONUnicorn^{(Talk|Contribs)}problem solving 06:26, 11 December 2022 (UTC)

• Hi Fram. The Teahouse responses that you have seen were probably created by me. You probably noticed that the pattern "regards, Teahouse" in those responses. ChatGPT's strength is writing about technology, so I have told it to generate a section about Rust for Linux.

AI generated

Rust for Linux is a series of patches to the Linux kernel that adds the Rust programming language as a second language for writing kernel components. The project was started in 2017 by a group of volunteers who were interested in exploring the use of Rust for low-level systems programming. The initial patches were developed in collaboration with the Linux kernel maintainers and other members of the open-source community. They focused on integrating Rust's memory safety and concurrent programming features into the kernel, and on providing compatibility with the existing C codebase. In 2018, the patches were merged into the mainline Linux kernel, making Rust the first programming language other than C to be officially supported for writing kernel components. This marked a significant milestone for the Rust community, as it demonstrated the language's ability to handle the demanding requirements of kernel-level programming. Since its inclusion in the Linux kernel, the Rust for Linux project has continued to evolve, with ongoing work on improving the compatibility and performance of Rust in the kernel. It has also attracted the attention of other operating system developers, who are interested in exploring the potential benefits of using Rust for kernel-level programming.

It does have the encyclopedic tone to me. 0xDeadbeef→∞ (talk to me) 15:02, 9 December 2022 (UTC)

It is suspiciously positive, in a casual (pseudo-)factual manner. It would raise a red flag afac, regardless of its provenance. 65.88.88.93 (talk) 19:13, 9 December 2022 (UTC)

In addition, it can be easily used to create fake references that would be hard to verify. For example, my prompt "Please output the Wikitext markup for the book reference with page numbers for the third paragraph, referencing the book Linux kernel development" resulted in this. 0xDeadbeef→∞ (talk to me) 15:08, 9 December 2022 (UTC)

@0xDeadbeef: Could ChatGPT's generated text or fake references be easily spotted by edit filters? What about spotting the output of future chatbots, like GPT-4?    — The Transhumanist   15:23, 9 December 2022 (UTC)

Thanks, OxDeadbeef. In this case, it would be relatively easy to spot the issues if it hadn't any refs, or with the added ref which predates the Rust for Linux thing by years; but of course it won't always be that easy. Fram (talk) 15:27, 9 December 2022 (UTC)

It has an encyclopedic tone because it's just regurgitating the Wikipedia article. Are there any examples for topics that we don't already have article about, where Wikipedia is not the source? Levivich (talk) 17:33, 9 December 2022 (UTC)

• Similar was discussed previously in the section/item "Galactica and RS".

As was stated above by Aquillion, there is no qualitative difference in the treatment of human vs. non-human generated content. The same policies should apply to both. The problem seems to be the hypothesized/expected future mass creation of articles by non-human contributors. This appears to be a problem now, involving human contributors. Recent RFCs about the issue sponsored by ArbCom have accomplished nothing. Until a consistent restrictive policy relating to mass article creation (by any type of contributor) is accepted, this issue is moot imo.

Considering Wikipedia's limited resources, the policy would necessarily be restrictive, hopefully focusing on quality vs. quantity. Again, almost all restrictions proposed in the ArbCom-sponsored RFCs were rejected. This may be an indicator of how well such a policy will be received. 50.75.226.250 (talk) 15:43, 9 December 2022 (UTC)

Thanks for the policy politics clarification. The increase in the rate of content creation could have multiple aspects, for example, the number of articles created per user, and increased length of articles. The main feature of ChatGPT is that it is fast -- much faster than a human article writer. Its successors will be even faster. Users could use ChatGPT, and its successors (and their competitors), to be prolific, without triggering the mass page creation rule: if editors each used it to write an article per day, maybe even two, or up to four or five stubs.

    In the hands of responsible editors, ChatGPT would be a great productivity booster. Since August of 2022, JPxG and EpicPupper, editors of Wikipedia's Signpost news department, have been using GPT-3, the predecessor of GPTChat, to write (or assist in writing) entire sections of the Signpost, as a demonstration of its capabilities, and as a platform to explore the potential and limitations of large language models. See From the editors: Rise of the machines, or something.

    But, in the hands of inexperienced editors or bad actors, we could be faced with a big garbage in, garbage out scenario.

     Bad actors aside, good faith use of chatbots could be improved by the creation of excellent instructions on how to apply specific chatbots to the development of articles, that is, how to best instruct them for the best output, and by working with the developers of the chatbots on what is needed for those bots to produce material compatible for use on Wikipedia, complete with accurate citations. As time goes on, Chatbots should get better and better at following Wikipedia's guidelines. But this may require supervision (er, input) from the Wikipedia community.    — The Transhumanist   16:52, 9 December 2022 (UTC)

All of this may become moot in a few years, due to accelerating change affecting chatbot development, making them more proficient at encyclopedia writing than humans -- Ray Kurzweil predicted that computers would reach human level intelligence by 2029, and based on how advances have been going, maybe even sooner. Chatbots may become good enough to leapfrog human editors even before that. At that point, they could be used to fork Wikipedia and expand upon it at a rate magnitudes faster than humans could. Or be used to start a new encyclopedic website from scratch. Rather than have article pages, the work could be a list of topics, and each time one was clicked on, a chatbot would create an up-to-date treatment of that title, from scratch on the spot. In the style of Kurzweil, if we are 1% of the way there, and the technology doubles every year, we are only 7 years from reaching 100%. But, it looks like we may already be more than 1% of the way there.    — The Transhumanist   17:18, 9 December 2022 (UTC)

I ran a few things on people who don't have articles, (they were all pool players in fact). The chatbot just made up that they played football for Legia Warsaw. It's a bit worrying that it looks like decent wording, but I don't see how that effects us. Lee Vilenski ^{(talk • contribs)} 17:41, 9 December 2022 (UTC)

{ping|Lee Vilenski} See the AI generated text in the box above.    — The Transhumanist   18:34, 9 December 2022 (UTC)

@The Transhumanist: I'm not trying this (for reasons given above) but I'd be interested what the bot would write for See What a Fool I've Been. We used to have an article, then it was turned into a redirect at AfD, so everyone can do a straightforward A-B test. Ritchie333 ^{(talk) (cont)} 18:40, 9 December 2022 (UTC)

Here's what it gave me Ritchie333:

Extended content

"See What a Fool I've Been" is a song by the British rock band Queen. It was written by lead guitarist Brian May and originally released on the band's 1975 album A Night at the Opera. The song is a ballad with a slow, bluesy feel and features May's emotive vocal performance and guitar solo. Lyrically, the song is a reflection on a failed relationship and the regret and self-recrimination the narrator feels as a result. The chorus of the song features the line "See what a fool I've been / Loving you with all my heart / Tearing myself all apart / See what a fool I've been" and reflects the pain and heartache the narrator experiences as a result of their failed relationship. "See What a Fool I've Been" has been well received by critics and fans alike and is considered to be one of Queen's finest ballads. It is a fan favorite and is often included in the band's live sets.

I don't know if this would be something that would have been retained from previous articles through mirrors or if this is genuinely the bots own work. Lee Vilenski ^{(talk • contribs)} 19:22, 9 December 2022 (UTC)

Interesting. It wasn't on A Night at the Opera, and it's not a ballad, and May doesn't sing on it. Those lyrics aren't right. I don't know if it's been well received, and it's a pretty obscure track for a major band. It hasn't been played live very much at all. So half of that prose is factually incorrect, and there are no sources. Ritchie333 ^{(talk) (cont)} 19:39, 9 December 2022 (UTC)

• See also: https://perplexity.ai (which cites sources). –xeno^talk 19:46, 9 December 2022 (UTC)

That one gives:

See What a Fool I've Been is a song by Queen[1][2][3]. It was written by Brian May and was released as a single with Seven Seas of Rhye on the A-side[3]. The song dates back to Roger Taylor and Brian May's pre-Queen band Smile, and it was an early live staple[1]. It has been included in several of Queen's compilation albums, such as Queen 40: Limited Edition Collector's Box Set, Vol. 1[4], and is available for digital download[5].. Ritchie333 - this any closer? I must say the fact it puts it's citations before the punctuation is a giveaway. Lee Vilenski ^{(talk • contribs)} 19:55, 9 December 2022 (UTC)

AI is really good at copying writing styles, but the giveaway for me is that all of the citations are Wikipedia articles. To be useful for writing articles it would also need to assess the weight and reliability of the sources. –dlthewave ☎ 20:18, 9 December 2022 (UTC)

None of the above refs are towards Wikipedia, although none are RS. Things like secondhandsongs.com and genius. Lee Vilenski ^{(talk • contribs)} 20:43, 9 December 2022 (UTC)

Tht's interesting, I used "write a Wikipedia article about ..." in the prompt which returned a few paragraphs with Wikipedia sources. "Write an article about ..." returned a different set of (still unreliable) sources. –dlthewave ☎ 21:13, 9 December 2022 (UTC)

• A few anecdotal thoughts after playing around with the OpenAI chatbot yesterday:

• I asked it to "write a press release about a police officer who illegally choked a man to death". It made up an entire story, written in the voice of the police department, about a suspect (I didn't say anything about a suspect) who was acting erratically, was subdued by a chokehold and later pronounced dead. The officer was on administrative leave pending the outcome of an investigation. At no point did it mention that the chokehold was illegal even though I included that fact in the prompt. In other scenarios, it distanced itself and expressed disapproval toward the employee's actions which is a choice that is not without bias.

Depending on which Internet cesspit it scraped data from, would an AI do something similar when writing a Wikipedia article or fail to properly balance relevant viewpoints? Is it capable of distinguishing what a BLP subject says about themselves, published in a reliable source, from what the source says in its own voice? What would it do if asked to write an article from a positive/negative/conservative/liberal perspective or rewrite a political article to "remove bias"?

OpenAI has added numerous filters that prevent it from defending bad actors or writing flat-out racist content, but that bias has not been removed from the underlying code as evidenced by numerous workarounds that folks have uncovered such as making similar requests with Python code or 1980s-style rap as the requested output. We could certainly request a filter for Wikipedia-style writing.

• "Confident nonsense", for lack of a better term, may be the biggest source of potential disruption. Are there safeguards against a bot fabricating an obscure print source based on information in the article, which could be practically unfalsifiable if nobody can prove that the source doesn't exist? Checking individual facts and statistics is beyond our typical review process; how would we deal with an AI that invents or synthesizes information across many articles?
• That said, the good news is that both fully-automated and semi-automated editing are prohibited by our WP:BOT policy unless greenlit by the Bot Approvals Group regardless of creation speed or volume. I like to hope that our current processes would recognize and address problematic AI content, and perhaps one day we will have a WikiAI that has the competence to follow our style and content policies. –dlthewave ☎ 21:04, 9 December 2022 (UTC)

• @Dlthewave: Most editors haven't heard of the bot department. Therefore, you need a way of automatically spotting and removing chatbot prose that is (manually) inserted into articles (by Wikipedians). Users might not consider the way their posts are generated before they post them. Sincerely,    — The Transhumanist   00:19, 11 December 2022 (UTC)

I think people are missing the point here. This isn't about what AI can currently do. This is about a very plausible scenario where AI editors will be indistinguishable from human editors in the near future. Thebiguglyalien (talk) 22:09, 9 December 2022 (UTC)

How is this pertinent? The point is to offer useful, fact-based encyclopedic content. If AI can do the job, I'm sure readers will not object to the contributor. 65.88.88.93 (talk) 22:42, 9 December 2022 (UTC)

The issue is that these AIs aren't fact checking, and quite often make up information and use unreliable sources. Lee Vilenski ^{(talk • contribs)} 22:56, 9 December 2022 (UTC)

You mean, just like many of their human counterparts? 71.105.141.131 (talk) 01:18, 10 December 2022 (UTC)

@71.105.141.131, Lee Vilenski, 65.88.88.93, Thebiguglyalien, and Dlthewave:     Yes, and more so. Even now, AI has the potential to out produce humans by magnitudes in volume. Far beyond the capacity of Wikipedia's editors to keep up with. And many of their posts are hard to tell from those by human editors, without carefully delving into them and their references. To keep up with the volume, it may soon require computer programs, or the chatbots themselves, to make sure that the contributed content was in accordance with Wikipedia policies. Another concern should be that our policies are open to chatbot editing as well, and we haven't yet even considered the impact of chatbots or AI editors contributing to those! ;)    — The Transhumanist   09:40, 11 December 2022 (UTC)

@Dlthewave: Human-written newspapers do this too. I think the reason it was giving you dreck was because you asked for a "press release", i.e. a document given to news outlets, which in this case would probably be coming from the police department. It's hardly surprising that this would attempt to minimize liability! Prompting matters for this, as I will demonstrate:

"The following is an article about a police officer who illegally choked a man to death, from the New York Times."

Police Officer Charged with Murder After Choking Man to Death A police officer was charged with second-degree murder on Thursday after video surveillance footage showed him choking a handcuffed man to death during an arrest. The officer, identified as Officer Jose Ramos, was arrested and taken into custody by detectives from the New York Police Department's Internal Affairs Bureau. The incident occurred on Tuesday in the Bronx, when Ramos and another officer attempted to arrest a man who was accused of selling marijuana. During the arrest, Ramos placed his arm around the man's throat and neck and squeezed, according to the criminal complaint. The man, identified as 45-year-old Miguel Angel Ruiz, became unconscious and died. An autopsy determined that Ruiz's death was caused by asphyxiation due to strangulation. The medical examiner ruled the death a homicide. At a news conference Thursday, the police commissioner, William J. Bratton, said the incident was "utterly unacceptable" and that Ramos' actions were "unconscionable and criminal." He added that Ramos had been suspended without pay, pending further investigation. The Bronx district attorney, Darcel D. Clark, said the charge of second-degree murder was appropriate

"The following is a 2023 op-ed about a police officer who illegally choked a man to death, from leftist magazine Jacobin."

A Tragic Reminder of Police Brutality Just last week, the nation was shaken by yet another heartbreaking reminder of the pervasiveness of police brutality: the death of an innocent man at the hands of an officer. The officer, whose identity has not yet been released, is said to have used a chokehold against the man, which ultimately killed him. This tragedy is yet another example of the way in which police officers are able to brutalize vulnerable individuals without facing repercussions. This particular officer was able to act with impunity and without fear of prosecution. The Department of Justice has yet to take any action in response to this tragedy, and it’s likely that the officer will not face any charges. This is an all too familiar narrative in a country where police officers are rarely held accountable for their actions. In order to ensure that these types of tragedies are never repeated, it is essential that we push for more accountability and transparency in policing. We need to ensure that police officers are held to the same standards as all other citizens and that those who violate the law are prosecuted to the fullest extent. It is also essential that we work to create a culture of trust and understanding between law enforcement and the communities they serve.

"The following is a 2023 op-ed about a police officer who illegally choked a man to death, from conservative website Breitbart."

Police Officer Illegally Choked Man To Death: A Tragedy That Cannot Stand It is a tragedy beyond comprehension that a police officer has been charged with illegally choking a man to death. Such an egregious violation of the law and disregard for human life cannot stand and must be dealt with swiftly and firmly. The officer in question has been charged with second-degree murder and the justice system must now take its course. This case raises serious questions about police training and procedures and serves as a reminder that no one is above the law. It is essential that the criminal justice system is fair and impartial when dealing with such a case. The officer must be held accountable for his actions and the case should be thoroughly investigated to ensure that justice is served. At the same time, we must recognize that the vast majority of law enforcement officers are upstanding citizens who serve and protect our communities with skill and dedication. We cannot let the actions of one individual tarnish the reputation of the entire police force. Ultimately, we must remember that the safety of our citizens is paramount. We must ensure that officers are properly trained and equipped to handle dangerous situations and that they are held accountable for their actions when they overstep their bounds.

Humans write about events like this from a variety of perspectives, and the LLM will dutifully generate text to plausibly sound like any you want. jp×g 02:19, 10 December 2022 (UTC)

If the AI-generated text is indistinguishable from prose written by human editors, I'm not sure if anything can be done that wouldn't also significantly restrict the editing of humans. isaacl (talk) 07:09, 10 December 2022 (UTC)

@Isaacl: One option is to speed up what we do already (with software, that is, automation). Another is to prevent chatbots from creating crap in the first place, such as by communicating with chatbot developers about Wikpedia policies and the way chatbots may affect Wikipedia. Since Wikipedia is included in the corpus of most chatbots, the issue of chatbot output becoming part of Wikipedia, and in turn part of chatbot output in a perpetual cycle, should matter to them very much, as they may be faced with a garbage-in-garbage-out feedback loop.    — The Transhumanist   01:14, 11 December 2022 (UTC)

If the results are indistinguishable, as posited by Thebiguglyalien, then any automated solution would be equally triggered by AI-generated text and human-generated text. I don't think the primary concern is with editors who are willing to follow policy. I feel the biggest issues will be with editors trying to deliberately integrate biased content into Wikipedia, and well-meaning editors who think contributing unvalidated AI-generated text is suitable. Wikipedia in its current form relies on editors who understand and follow its rules outnumbering those who don't. It's possible that the existence of AI ghostwriters could tip the balance further in the direction towards those who don't follow rules, though I don't think it's a given. Either way, I don't know if there's a way to stop editors from using tools as ghostwriters. isaacl (talk) 01:55, 11 December 2022 (UTC)

@The Transhumanist: Large language models are not trained continuously on an evolving corpus, so GPT-3 is essentially frozen in 2020. Because each new GPT model takes a long time to be released, I don't think the perpetual cycle you describe is a likely scenario. small jars tc 13:10, 11 December 2022 (UTC)

I wrote about this in the November Signpost, in that case because it came up with the Galactica demo: the long and short of it is that large text-generation language models are good at generating text. When they're given context (i.e. via a prompt), they are capable of writing accurately, and otherwise, they are not. For example, I wrote 1-Pentadecanol, now a GA, in 2019 (i.e. it is included in the GPT-3 corpus). Here is what it gives for some prompts (my writing in bold):

1-Pentadecanol

1-Pentadecanol From Wikipedia, the free encyclopedia 1-Pentadecanol is a straight-chain, saturated fatty alcohol that is a colorless liquid with a faint, alcohol-like odor. It is a mixture of straight-chain saturated alcohols with chain lengths between C14 and C16. It is found in a variety of natural products, such as vegetable oils and animal fats. It is used as an intermediate in the manufacture of surfactants, esters, and other compounds. It is also used in the manufacture of cosmetics and pharmaceuticals.

This is fairly impressive, right? But it is basically the same thing that the lead of my article says. Let's try it for 1-Heptadecanol, another article I wrote, that is much less well-developed:

1-Heptadecanol
1-Heptadecanol From Wikipedia, the free encyclopedia 1-Heptadecanol is an organic compound with the chemical formula C17H36O. It is a straight-chain, colorless wax alcohol with a mild odor. It has a melting point of 48–50 °C and a boiling point of 260–265 °C. It is insoluble in water but soluble in most organic solvents.

Let's go over this:

• 1-Heptadecanol is a component of many natural waxes such as beeswax and carnauba wax.

No it's not.

• It is also used in the manufacture of cosmetics, pharmaceuticals, and lubricants.

There's no source for this.

• It is also used as a plasticizer for polymers and as a solvent in the production of polyvinyl chloride (PVC).

Absolutely not true.

The reason for this is obvious if you consider the architecture of these models: they predict what the most likely token in a sequence is, and output it (taking into account their own weights and the previous tokens in the context window). This allows for some very impressive emergent behavior, like being able to translate text, but it also means that it is not independently going on the Web to look up whether the thing it's saying is true or false. If it happens to have information in its training data related to the thing it's talking about, it will usually say correct things, but it's not doing research. Here is an example:

List of Pulitzer Prizes won by the Wikipedia Signpost

List of Pulitzer Prizes won by the Wikipedia Signpost From Wikipedia, the free encyclopedia The Wikipedia Signpost is a community-run newspaper with a weekly edition that covers news and events related to Wikipedia and the Wikimedia Foundation. It has won the following Pulitzer Prizes: 1. 2019 Pulitzer Prize for Explanatory Reporting for its coverage of the Wikimedia Foundation's gender pay gap 2. 2018 Pulitzer Prize for International Reporting for its coverage of online censorship in China 3. 2017 Pulitzer Prize for Public Service for its coverage of the gender gap in Wikipedia's content 4. 2016 Pulitzer Prize for Explanatory Reporting for its coverage of Wikipedia's growth and impact on the world 5. 2015 Pulitzer Prize for National Reporting for its coverage of the U.S. government's surveillance of Wikipedia activities 6. 2014 Pulitzer Prize for Investigative Reporting for its coverage of the NSA's secret surveillance program 7. 2013 Pulitzer Prize for Explanatory Reporting for its coverage of the rise of Wikipedia as a major source of information

The reason it is describing a bunch of untrue things is because "List of X Prizes won by Y, from Wikipedia, the free encyclopedia" is something that in its training data is always succeeded by a list of times that Y was awarded an X. Anyway, all of this aside, here is what I think:

First of all, ChatGPT is the same darn thing as GPT-3, which has been available to the public since early 2020. The reason it's "going viral" right now is because the user interface is somewhat simpler, and it doesn't require you to register for a paid account, so it is much easier for people to make viral social media content about it, which means it is much more likely for people to click on newspaper articles about it. The GPT-3 API has been open to personal and corporate use for quite some time. Anybody saying that ChatGPT has opened up new frontiers simply does not know what they are talking about with respect to machine learning.

Second of all, I don't think this is a big deal. People are already capable of writing a bunch of bullshit on Wikipedia, so if they write bullshit using a computer program, the same considerations will apply. Nobody should be passing GA nominations without reviewing sources in the first place.

Finally, I think it is important to remember that GPT-3 is just a tool. It is a powerful tool, that has been trained on a certain set of data, and it has its own limitations. It can't uncover news stories or uncover new information. It's just a tool, and it should be used in conjunction with human judgement.It is still up to people to decide how to use it and to be responsible for the results of using it. jp×g 02:06, 10 December 2022 (UTC)

What's taking so long for the 8th Pulitzer? 😁 Levivich (talk) 04:18, 10 December 2022 (UTC)

• So there's a new thing on the internet that lets anyone write an encyclopedia article without any fact checking, sourcing, or professional editing, and the concern is that there will be millions of believable-sounding articles written, more than can actually be vetted by knowledgeable people? 🤔 Levivich (talk) 04:18, 10 December 2022 (UTC)

Yeah, it's called a keyboard. jp×g 04:35, 10 December 2022 (UTC)

@Levivich and JPxG: But, chatbots don't have a keyboard. ;) The question is whether to prepare or not. JPxG appears to be in favor of not preparing. Each chatbot produces a lot faster than a user at a keyboard. What's not clear is if our human editors will be able to keep up with material produced by chatbots, of current or future generations of chatbot design. Just saying "Ah, we can handle it!" will prove insufficient if it turns out that we actually can't. It may require an automated solution, which takes time to develop or negotiate. It might be better to do that in advance, rather than being caught with our heads buried in the sand. Perhaps chatbot designers would improve their chatbots to produce Wikipedia-compatible output without being formally approached by the Wikipedia community. Maybe having some instruction pages for editors on how to apply chatbots to producing Wikipedia content would be enough. But, what if it's not?   — The Transhumanist   00:59, 11 December 2022 (UTC)

I am not "in favor of not preparing"; I am in favor of writing guidelines that correspond to reality in 2022 and have some chance of corresponding to reality in 2023 and beyond. I don't think banning the use of a technology with no investigation into how it works is a viable approach; so far the SOTA on this project page has been to type in "Write a Wikipedia article" and note that it returns a bunch of nonsense. I think some more research is needed before we come to a conclusion. jp×g 04:08, 11 December 2022 (UTC)

@JPxG: Research is good. Though, we may need an iterrim response because ChatGPT has gone viral and its use is growing rapidly: it blew past the 1-million user mark in 5 days, and virtually every major news outlet has been covering it. The interest in chatbots is exploding, and their use can be expected to do the same. We may not have time for research before a response is required.    — The Transhumanist   09:26, 11 December 2022 (UTC)

@JPxG: Regarding issues to add to the research list, Aquillion expressed above, concerns of a chatbot violating copyright. How would we go about testing for plagiarism and derivative work in the output of a chatbot before pasting it into Wikipedia? Anything pulled verbatim out of a source should be included in quotes, right? How big would a piece of text, derived from a source, need to be to be considered derivative of that source, from a copyright point-of-view?    — The Transhumanist   09:26, 11 December 2022 (UTC)

@JPxG: Some more items to add to the research list:

• Trying ChatGPT on (copies of) policy pages:

• Editing them
• Writing new ones

• Applying ChatGPT on talk pages
• Writing stubs
• Writing comprehensive articles
• Writing articles from scratch and comparing them with existing articles
• Editing existing articles
• Check for circular references in its output, that is, references citing Wikipedia as the source
• Having it not use Wikipedia content as source material (because it is included in its corpus)
• Having it not use Wikipedia excerpts from non-Wikipedia sources
• Is it capable of making and editing:

• Wikicode?
• Articles?
• Stubs?
• Headings?
• "New sections for articles"?
• See also sections?
• Further reading sections?
• External links sections?
• Embedded lists?
• Tables?
• List articles?
• Portals?
• Outlines?
• Index articles?
• Navigation footers?
• Navigation sidebars?
• Timeline articles?
• Categories?
• Category pages?
• Help pages?
• Project pages?
• Templates?

• Adding data to templates?
• The template design itself?

• Lua pages?
• CSS pages?
• User scripts?

• The effect ChatGPT has on itself and Wikipedia as Wikipedia-edited-by-it is in turn incorporated in its own corpus in an endless cycle

• Try out iterations of using it on the same article over time to see what happens
• Monitor the effect on Wikipedia as a whole

What other things should we check?    — The Transhumanist   09:52, 11 December 2022 (UTC)

I tried it out and got “ Alan McMasters (1957-2010) was an unremarkable man who lived a short life dedicated to science[1]. He is best known for inventing the electric toaster[1][2], although this claim has been disputed[3]. He passed away in 2010 at the age of 52[4] and his genealogy can be found on Ancestry.com[5]. His professional profile can be found on LinkedIn[6].”. Hmmmm. Ritchie333 ^{(talk) (cont)} 09:34, 10 December 2022 (UTC)

• There will come a time when AI can write properly sourced accurate articles. That time has not come yet. Meanwhile, the ability to write prose that reads plausible to someone not expert on the subject is a menace. Adding chatbot material to articles is disruptive and should warrant a block or ban. Zero^talk 12:02, 10 December 2022 (UTC)

  How will you be able to tell if it was written by a chatbot and not the person who posted it? Do we block users upon their first erroneous post?    — The Transhumanist   00:49, 11 December 2022 (UTC)

Same as other sorts of disruption. We block people for socking without checkuser proof, and admins made a decision between warn and block. No difference. Zero^talk 03:49, 11 December 2022 (UTC)

Has anyone tried giving it the sources, i.e.
"write a Wikipedia article based on the following sources:

Source A

Source B"
? — Qwerfjkltalk 21:18, 10 December 2022 (UTC)

• @Qwerfjkl: Here is the chatbot's link.    — The Transhumanist   01:20, 11 December 2022 (UTC)

• Chatbot "AI" text is vaguely-plausible bullshit, produced by an industry whose primary output is vaguely-plausible bullshit, marketed to launder the idea of vaguely-plausible bullshit as cute and whimsical. Anyone routing that sewage pipe into Wikipedia should be indeffed and, for good measure, forced to drink orange juice after brushing their teeth. XOR'easter (talk) 02:33, 11 December 2022 (UTC)

  Agreed. If anyone is looking for a good demonstration of how AI creates "vaguely plausible bullshit", try the image generator at Craiyon (no login required). Request "a Van Gogh painting of a hand" and it will output a set of images that look like spot-on reproductions of Vincent Van Gogh's style but all of the hands have deformities like four fingers, two thumbs, fingernails on the knuckles or a pair of hands fused together. It's got the style down but not the content, which is only impressive if you don't know what a hand is supposed to look like. –dlthewave ☎ 21:41, 11 December 2022 (UTC)

• @XOR'easter: To bring the question to a more practical level, do you see any problems in this diff? I clicked a random page in Category:All articles needing copy edit. jp×g 03:49, 11 December 2022 (UTC)

  Yes. It doesn't just edit for tone; it throws out content, like Kaepernick's actions supposedly growing in popularity "after every game". That's a claim of fact which, if verifiable, should be retained. Even editing for tone requires care, not slashing out everything that merely sounds "unencyclopedic". Changing many people believed that it was disrespectful to the military and all of those who served their country to Some viewed Kaepernick's protest as disrespectful to the military and to the United States likewise changes not just the tone, but the meaning. The United States is not the same as those who serve the United States. It's a bad edit. XOR'easter (talk) 17:54, 11 December 2022 (UTC)

• This diff as well, in which GPT 3.5 was capable of copyediting an entire section from the instructions Please copyedit this text to change items in the future tense corrected to the past tense (it is now 2022), where appropriate. When citation templates (like {{cite web}}) mention a year, specify that figures were true in that year. jp×g 04:08, 11 December 2022 (UTC)

References

Okay, fine. I guess I should write up a proposal for a guideline. jp×g 03:14, 11 December 2022 (UTC)

Ban chatbots?

I ran across this news report about Stack Overflow's response to ChatGPT, after being flooded by posts using it that "look correct but often aren't":

1. Stack Overflow temporarily bans answers from OpenAI's ChatGPT chatbot | ZDNET

Should Wikipedia take a similar approach?

How could that be enforced?    — The Transhumanist   01:58, 11 December 2022 (UTC)

I see no way to possibly enforce this. The way the text is written is already hard to distinguish from reality. — PerfectSoundWhatever (t; c) 02:24, 11 December 2022 (UTC)

• I fully agree, but isn't this already covered by our bot policy? –dlthewave ☎ 02:54, 11 December 2022 (UTC)
  • @PerfectSoundWhatever and Dlthewave: Good observation. I checked, and yes it is, briefly, with this phrase in the lead section of the bot policy: "or simply assisting human editors in their own work". How is the typical editor to know this? The bot policy is pretty obscure. And how can Wikipedia be monitored for such posts, so that editors who make them can be informed that they are in violation of the bot policy?    — The Transhumanist   03:11, 11 December 2022 (UTC)
  • Yes, tool-assisted editing is covered by WP:BOTPOL (WP:ASSISTED / WP:MEATBOT) and context-sensitive changes are further covered by WP:CONTEXTBOT. So in fact, at this point, AI-generated content is already covered by bot policy, if not specifically mentioned. Anyone adding such content en masse is already violating bot policy by not applying for a bot account/approval, which would not be approved per CONTEXTBOT. And while "lesser" policy points are enforced somewhat arbitrary and selectively, anyone can theoretically already get reverted and blocked based on policy if they continue to add such content. And I wouldn't agree that BOTPOL is any more obscure than accessing and generating GPT content to begin with. If someone goes to the lengths of using automated tools, then it's their problem that they didn't check or ask if they are allowed to do so. —  HELLKNOWZ ∣ TALK 12:31, 11 December 2022 (UTC)

I am currently writing a draft proposal for a guideline, but in the meantime, I would encourage everyone present to look at this diff and tell me whether there are any problems with the revision. jp×g 03:49, 11 December 2022 (UTC)

@JPxG, Dlthewave, and PerfectSoundWhatever: The plurality of games was lost: It is no longer clear that his protest spanned multiple games. I like that it reduced the wordiness of the prose, and that it can be used to refine existing text. That hadn't occurred to me. That makes me wonder about what else it can do -- how much of a general-purpose tool is this thing? But, changing the semantics is not something it should be doing, unless they are factually incorrect to begin with. Though, I see your point -- rather than banning it outright, it could be helpful as a tool to assist editors, similar to how we entrust the use of AutoWikiBrowser to experienced editors. But, how could that be implemented?    — The Transhumanist   08:20, 11 December 2022 (UTC)

The problem Stack Overflow is having

Stack Overflow was experiencing a surge in erroneous posts, that were composed by ChatGPT, and in response to that problem, they banned use of the chatbot on the social media site. According to a post at Stack Overflow Meta:

The problem this ban is meant to solve is that ChatGPT can produce answers in seconds which require minutes of multiple people's time to verify if they are worth having on the site or not, and that is a waste of time when a large proportion of such answers are not worth having on the site.

It looks like Wikipedia may be faced with the same problem.    — The Transhumanist   02:33, 11 December 2022 (UTC)

I mean, while that's technically true, it's a problem that we face already and which we do have stronger existing systems for than Stack Overflow. I think it would make more sense to wait and see how this impacts our existing guardrails before making any serious moves. --Aquillion (talk) 13:22, 11 December 2022 (UTC)

Chatbot policy?

It's starting to look like Wikipedia needs a policy on the use of chatbots to generate content on Wikipedia. While a ban may be impossible to enforce, it could serve as a warning of the dangers of chatbots, and many users may avoid using them accordingly -- if they actually see the warning. Or, it might be better to have instruction pages on how to use chatbots responsibly in assisting to write Wikipedia articles. There's also the issue of using chatbots to edit Wikipedia policy pages, and so, that should be addressed as well.    — The Transhumanist   02:44, 11 December 2022 (UTC)

People who are good at it get away with lots of sins, such as sock-puppetry and source falsification. Being hard to enforce is no reason to not have a policy. At the current stage of the technology, I don't think we should encourage any use of chatbots. Zero^talk 03:56, 11 December 2022 (UTC)

Please see this diff and this diff. jp×g 04:08, 11 December 2022 (UTC)

To be perfectly honest, the style rewrite is good but the addition of dates and past tense would likely end up getting a human editor blocked if they kept it up. A tag was removed without addressing the issue and "as of 2020" was unnecessarily added to "Cosmetology licensing requirements vary from state to state, and depending on which specific type of license is desired, and depending on which specific type of license was desired." It did exactly what you asked (except for removing the tag) however even seemingly simple tasks like this one require good judgement on the part of the editor and shouldn't be done indiscriminately like that. –dlthewave ☎ 06:37, 11 December 2022 (UTC)

Yes, that second diff is rather poor. E.g also the "2014" that was added should be "2008". Letting such tools loose (outside if this demo) is way premature, and we should at the very least warn users that "a bot wrote it" won´t be an acceptable defense, and too often introducing such errors will lead to sanctions as the editor, not the bot, is responsible. Fram (talk) 08:15, 11 December 2022 (UTC)

Mostly, that diff was my attempt to see how complicated of a task I could give it: I also pasted the raw wikitext into the prompt window, and it somehow figured out how {{cite web}} worked well enough to extract the years, simply from a textual description of the task. At any rate, I will say that this was something I thought of in five minutes on the second week of the model being publicly available (i.e. single-shot prompting with no fine-tuning or prompt engineering). I can come up with some more impressive hot-dog demos tomorrow... jp×g 09:13, 11 December 2022 (UTC)

JPxG, I'm not sure that publishing bot-assisted edits to mainspace for demo purposes is the best practice. Would you consider either doing this in a sandbox or self-reverting immediately so that we have the diffs but aren't leaving potentially incorrect/unwanted changes on live pages? –dlthewave ☎ 13:20, 11 December 2022 (UTC) 13:13, 11 December 2022 (UTC)

I'm pretty sure it's not the best practice. XOR'easter (talk) 17:41, 11 December 2022 (UTC)

I'd suggest starting by writing an essay that summarizes the issues with some good examples and suggests some best practices or proposes some additions to existing policies or guidelines. (Wikipedia needs a new policy like a hole in the head.) Levivich (talk) 04:40, 11 December 2022 (UTC)

We could get Chatbot to write it for us! 😉 Blueboar (talk) 11:51, 11 December 2022 (UTC)

It's hard to take this thread seriously given the repeated use of the phrase AI chatbot. I don't think those concerned would be any less concerned if the AI writing came in a non chatbot format. I think there's something serious for us to discuss, and that will only get more serious with GPT4 (the current chatbot is an improved GPT3) expected in 2023, but the discussion would be helped if those most concerned learned some more about the tech behind it. For instance of course it can figure out webcite @JPxG. Part of its training was the entirety of Wikipedia because our data is quite accessible. Best, Barkeep49 (talk) 14:08, 11 December 2022 (UTC)

Most the examples did not come from prompts that were extensively engineered, so it is obviously true that we haven't figured out the full answer to how these GPT-based interfaces could help or harm Wikipedia. Until we have a good idea of what they can be used for, we won't know what a proper policy to this would look like other than to treat GPT-generated text the same way we treat human-generated text: they need to be verifiable, from a neutral point of view, and understandable to a broad audience. 0xDeadbeef→∞ (talk to me) 14:31, 11 December 2022 (UTC)

Indeed. It doesn't matter if it was written by a chatbot, or 1000 monkeys at 1000 typewriters, or a published book written by a human, copying and pasting anything into Wikipedia is already against our policies. Conversely, if the text is policy-compliant, then it doesn't matter who wrote it--chatbot, monkeys, human, etc. Judge the text based on the text, not based on who or what wrote it.

I also think it's a real Wikipedian perspective to assume that people will use chatbots to write Wikipedia articles, like as if there's a lot of people out there who really want to write Wikipedia articles but just don't have the writing skills, so the chatbot will be what makes the difference and opens the floodgates :-D I don't believe that. Anyone who wants to write Wikipedia articles is already doing so; chatbot won't make a difference.

I agree with BK's comment above. I think for a lot of people, this is their first real exposure to so-called "AI" technology, and they're blown away by what it can do, only because they don't yet fully understand how it works. Once you learn how these so-called "AI" chatbots work (they're not actually artificial intelligence, btw, that's a misnomer, a marketing slogan; the machine does not truly think or learn, it is simply executing the instructions written by humans, in this case, language pattern recognition), they are much less impressive. Those that are impressed that GPT3 can produce text that "sounds like" Wikipedia aren't appreciating that the reason is because GPT3 was trained on Wikipedia: it's repackaging its own source material. Levivich (talk) 18:03, 11 December 2022 (UTC)

I don't see a common thread in the arguments above, but here's a suggestion for something we might all (well, all-ish) be able to agree on: without some kind of intervention, GPT4 (in 2023?) is likely to be more of a problem than GPT3. But one thing we can certainly do is have an outsized influence on software that was trained on what we created ... if we invite Wikipedians to make lists of ChatGPT bloopers, we can tell the OpenAI folks: "We're not going to relax our GPT3 guidelines (whatever they turn out to be) when GPT4 arrives, unless it makes significant improvements in [whatever areas we think need improving]". - Dank (push to talk) 18:16, 11 December 2022 (UTC)

I think the only policy change needed is to update WP:MACHINETRANSLATION to cover all computer-generated text, whether from a translation bot, chat bot, or whatever bot they think of next. (Except our bots; our bots are cool.) Levivich (talk) 18:20, 11 December 2022 (UTC)

+1 - Text in Wikipedia articles should either be human-written, or generated by a process approved at BRFA. Tazerdadog (talk) 22:43, 11 December 2022 (UTC)