Discworld & Pratchett Wiki:Mended Drum

 This is a location to discuss non-content matters (what do we do with content disputes, vandalism, etc, what do we want to do with this wiki, and so on).

This is the page for current discussions.

Archives

• Discworld & Pratchett Wiki:Mended Drum/Archive 1
• Discworld & Pratchett Wiki:Mended Drum/Archive 2
• Discworld & Pratchett Wiki:Mended Drum/Archive 3
• Discworld & Pratchett Wiki:Mended Drum/Archive 4
• Discworld & Pratchett Wiki:Mended Drum/Archive 5
• Discworld & Pratchett Wiki:Mended Drum/Archive 6
• Discworld & Pratchett Wiki:Mended Drum/Archive 7
• Discworld & Pratchett Wiki:Mended Drum/Archive 8

AI content and fan project promotion

So we have a new contributor (hello DugBride if you’re reading this) who is adding AI-generated images to pages without images, and attributing them to their fan project, a Discworld “re-skin” of the board game Oath. Which means the time has come for us to have a policy about AI-generated content. As a writer, actor, teacher, podcaster and more, every facet of my life is currently being made worse or at best more complicated by diffusion model images and large language model text, so I’m not a fan. But I recognise that’s not necessarily a majority opinion. So: what do we think? I don’t think we want AI text here at all, and at a bare minimum I want art to be clearly attributed to the model that created it. I would prefer hand-drawn fan art for character and item art, and I confess I am not a fan of some of the images uploaded so far, but again I recognise that’s a preference. But I am also not in favour of the wiki being filled up with images that promote a fan project, even if it is one I’m interested in. (This is why I have limited mention of my own podcast here, and started my own wiki for the detail I wanted to share.) What do you folks think? -- Guybrush (talk) 22:48, 7 December 2023 (UTC)

I'm not a fan either, of course, and I would certainly ban AI-generated text, except how would you recognise it? Images are more difficult. What if the image were created in an old-fashioned (two years ago?) drawing program? What about photographs? I expect you know a lot more about the process than I; what if the AI is just used to smooth out an original drawing or color it, or...? Enforcement seems to be the hardest part. Other ideas, anyone? Old Dickens (talk) 23:37, 7 December 2023 (UTC)

I’m not proposing we get too draconian about it, but having an official policy (or as close to official as we get) is enough to get us started. If folks want to ignore the rules and do it anyway, we can address that if necessary. I would just like us to be clear about whether we want images created by a diffusion model - which is to say, a generative AI like MidJourney or Dall-E which takes a text prompt and then creates an image based on analysis of training images (many used without permission) and associated descriptions. If someone makes art themselves and includes some kind of computer assistance in their process, I am not that fussed, just as I don’t mind if someone has a chat with ChatGPT to get inspiration for the fanfic they then go on to write. But the kind of thing created just by prompting an algorithm seems rather against the fandom spirit of L-Space, to me; fandom involves community, and that means if you need art for a project, you talk to other fans and find fan artists who are willing to help. An example policy might be something like: “This wiki does not permit the use of content created wholly or primarily by generative AI systems, including large language and diffusion models for text or art. All text added should be the work of the contributor; any art uploaded should be done so only with permission, and attributed to the creator.” -- Guybrush (talk) 08:00, 9 December 2023 (UTC)

Hi there thanks for the welcome and open discussion I was aware both issues you have raised were contentious which is why I messaged Old Dickens direct before touching a page to get the go ahead to open this can of worms. I don't know how the back end of these things work so had to feel my way and have learned a fair bit in the past couple of days. Including where you talk about these things.
Here is where I am at...I would say if you don't like AI content then don't allow any AI content, if it is uploaded but then attributed to the model then I think you would be promoting it and will get a lot more. It has been surprisingly easy for me as a first time user to access and upload images seemingly without any gatekeeping, I am sure you already have safeguards against this but you may want to relook at those if you are hoping people will act in good faith. As for the project promotion, I agree it's a bit too much and if it's OK I will remove all references just leave the link in in my profile. Following the example set by Guybrush. I will of course not add any more pics but, it might be best if the current images stay while you make your decision, a picture tells a 1000 words even if those words are “we don't want these pictures” if then they are removed you may have to do it because although I can remove the page code, I have no idea how to delete uploaded files. Ironically I have been using both this reference and Guybrush’s podcast regularly without knowing they are connected for the last few months to get up to speed on Discworld, it took me all my life to read most of the books and so I have had to take shortcuts to work on the project references. I have a lot of respect for both sources and wouldn't want to cause any trouble.
If the images are unwelcome then I doubt I will have much to add as a contributor, there are a few unanswered questions when doing my research that I might be able to offer something on; Holy horns gesture? Guild Weathervanes? But you seem to have a lot covered and I have my own shed to be working in, maybe further down the road if you need some help administrating incoming imagery then I can help, although my field is Digital Imagery I am no artist and no expert in spotting deep fakes either. Finally I am thrilled you have both looked at my work in progress even if your not fans, I appreciate discerning feedback but happy to keep in my shed for now where I can choose how long Carrot has grown his hair. DugBride 11:44, 9 December 2023 (UTC)

Hey DugBride - I might not be a fan of the art, or even Oath, but your project is pretty amazing! And we do have a Fandom page where you could list it, at the very least. I want to thank you for engaging in this discussion in generous good faith, too. I’m not interested in having to police images really, but I think if we have a policy and find the right place to get it in front of users then that should be enough - I’m not expecting a torrent of diffusion model generated Discworld characters any time soon! — Guybrush (talk) 20:10, 14 December 2023 (UTC)

I'd go along with Guybrush's policy statement, above, now, where does it go? I wouldn't like a big banner of "don'ts" on the front page, but if it's in "About", or Help:Editing nobody will read it. Help:Editing is already noted at the top of every editing page, so I guess we could say ignorance of the law is no excuse. I have to defer to Guybrush on interpretation of what's too much AI; I don't suppose it's possible to credit the source in an AI image when there might be dozens. Old Dickens (talk) 18:44, 10 December 2023 (UTC)

I think on image upload I had to tick a box marked "this is my own work" If it had said this is my own work and not AI generated I would have halted at that point. Apologies for the multi edits; still learning. --DugBride (talk) 12:33, 11 December 2023 (UTC)

This is a pretty good idea. What if we create a policies page (it could also include guidelines on annotations etc) and then link to that from the upload page? We can probably change the language on the upload page itself, too. And to be honest ownership is part of the problem; at the moment the ownership of generated images is murky at best, and it’s certainly unclear who owns copyright. So that does seem like the right place to put it. — Guybrush (talk) 20:10, 14 December 2023 (UTC) On the subject of “how much AI is too much AI” I don’t think that’s a big problem; right now the thing we don’t want are wholly generated images - things created by diffusion models and similar tools, stuff like MidJourney or Dall-E, which create an image for you based on a text prompt. We don’t need to worry about folks using “AI” tools when making digital fan art (anything vaguely automated is being misleadingly lLabelled “AI” At the moment, even when many of those technologies already existed under other more sensible names). Perhaps we just need a clarifying clause on the “your own artwork” covering that it has to be yours, used with permission, allowed by copyright law and attributable to a person. I think it will also help to include in our art policy that as a fan project we want to showcase fan artists; you are welcome to use whatever you like to make art for yourself or or your own projects, but here we want to showcase human-drawn (etc) fan art. And as a longer term solution, perhaps I can do some call-outs in other fan spaces asking if folks have character and location fan art they’d be happy for us to use on the wiki for articles which lack images? — Guybrush (talk) 20:24, 14 December 2023 (UTC)

I second all that, except that the Upload file page is for images and doesn't relate to annotations; also it's a "special page" and Osiris would have to modify it. Help:editing is already linked from editing pages. Old Dickens (talk) 21:19, 14 December 2023 (UTC)

DugBride has removed his contributions. I'm of two minds on this; I have despised computer-generated pop music for many years now and I would prefer hand-made art or clever photographs, but we've begged for illustration for years and some of what we have isn't very good. I wish we could have more as useful as PetuliaGristle.jpg for example. Are these opposite and irreconcilable viewpoints? Old Dickens (talk) 04:17, 16 December 2023 (UTC)

Missed one, just removed Detritus from the page this morning, like I said I can't do much about the images I uploaded as it won't let me delete the copy on the server. I sympathise with your issue, there are significant characters and locations without graphical representation. The AI works for me as I have to produce 250 images for a very small audience on zero budget, and I am more keen on design and mechanics than artistry. I don't think its as vital a part of your offering. Its possible that in the new year I will take Guybrush's advise and reach out to some of the fan-artists to see if they want to help my project but only when I have something significant to demonstrate. I'll be happy to mention the gaps here that need filling, but it sounds like you have already been down that road. --DugBride (talk) 12:00, 18 December 2023 (UTC)

Hi, all, sorry I'm late to the discussion... boarding school at Hogswatch is a time sink. Anyway, y'all have said pretty much everything I would have, so all I'll do here is lend one more voice in support of Guybrush's suggestion, This wiki does not permit the use of content created wholly or primarily by generative AI systems, including large language and diffusion models for text or art. All text added should be the work of the contributor; any art uploaded should be done so only with permission, and attributed to the creator. I don't think we need to do much more in terms of advertising or enforcement, as long as we have something like this clearly stated somewhere. People don't necessarily read rules before editing. If we find someone using ChatGPT to make articles, we politely ask them to stop, and can point them to the statement if they wonder whether the polite ask has community consensus behind it. Moishe Rosenbaum (talk) 01:01, 21 December 2023 (UTC)

Have you watched The Holdovers? Old Dickens (talk) 02:30, 21 December 2023 (UTC)

I haven't, but I just read about the movie... boarding schools are (for the most part) very, very different places in 2023 than they were in 1970. Yet, the popular perception of class privilege and influence lives on in the zeitgeist. My school doesn't allow students to remain on campus over Christmas break. The ones for whom getting home would be prohibitive - usually just a few folks from overseas - generally stay with friends. We have a faculty member who helps coordinate such stays if students need help. My family hosted several of my son's friends, who were from China and Vietnam, on occasion.

Okay, in spring 2020 most of our international students remained on campus, and I was one of many faculty who volunteered to chaperone / entertain them. That's a very different story, with very different cultural context, to what I read about The Holdovers. <smile>

Atom

Clicking Atom in the sidebar bar produces the message: "This XML file does not appear to have any style information associated with it. The document tree is shown below.", followed by pages of XML gibberish. Not widely useful. Old Dickens (talk) 03:34, 26 December 2023 (UTC)

The Atom link is only visible in Recent Changes, your Watchlist or the history of a specific page, and generates an RSS feed (in Atom format) from entries in that list using XML which is (hypothetically) useful for other apps or websites to read. The main use case I’ve seen for that recently is a Discord bot which posts recent changes to a specific text channel, but generally speaking RSS has fallen from favour for most uses except podcasting. The Atom feed option is built in to MediaWiki so we can’t remove it by editing the sidebar menu, but probably there’s an option to disable it in the config file if we really want to. (FYI I don’t get the error you mentioned, though I did get a different one trying to view the feed for my Watchlist that suggests we might have an out of date extension somewhere?) — Guybrush (talk) 12:29, 26 December 2023 (UTC)

Spamagain

Suddenly, the SEOs are back! Old Dickens (talk) 16:34, 14 March 2024 (UTC)

Might be time to turn on new account approval? I run a plugin for that on my wiki, works pretty well. -- Guybrush (talk) 09:08, 12 June 2024 (UTC)

May I ask the name of the plugin you use? I wanted to switch to Turnstyle captcha but that isn't in the stable release of Mediawiki yet --Osiris (talk) 04:02, 13 June 2024 (UTC)

I use Confirm User Accounts. I get a little spam (I just checked and there were three junk requests) but they doesn’t slow down the wiki or result in any vandalism. It has a “Spam” response which doesn’t send an email to the requester. The main downsides are that you have to respond to each request individually, and it does throw a couple of error messages, though nothing serious. (I’m currently running MediaWiki 1.40.1 and Confirm User Accounts 224079f.) It might not be the best available! -- Guybrush (talk) 07:44, 14 June 2024 (UTC)

Twenty-five at once is getting back to the bad old days! Old Dickens (talk) 16:01, 1 August 2024 (UTC)
Forty-three so far today, almost all from the same source. I wonder what the "human not bot" filter does. Old Dickens (talk) 16:25, 14 August 2024 (UTC)

It's getting to be a bit much, and for extra annoyance, when I'm logged in I can't see any but the most recent deletions. If I log out, I can see that the spammers I've just deleted still show, but I can't do anything with them. Fortunately, they don't seem to be active but they still show in Recent changes. Can AI-assisted bots now defeat any captcha? , Old Dickens (talk) 02:27, 11 October 2024 (UTC)

The Chicago insurance agent is getting really tiresome. It can do fifteen at once all day and night, apparently automated. Old Dickens (talk) 18:13, 4 November 2024 (UTC)