Re: Wikipedia at 25: A Wake-Up Call (essay)
Kim Bruning via Wikimedia-l <[email protected]> Sat, 2 May 2026 14:17:59 +0200
| Newsgroups | gmane.org.wikimedia.foundation |
|---|---|
| Message-ID | <afXrd0c5ppI3Ia0q@escher> |
Anders Wrote: > *articles where AI gets better result than our articles, (and here have to > think through how to react, change our behavior) Are you thinking of foundation models? Or pure chat interfaces. AI in a box? "You're absolutely right!! ✨ That's from past my knowledge cutoff, I'm so sorry 🥺 I'll try again better next time :-) 💕" [1] Aka: your infamous AI slop approach. If you use a 2026 agent with a bunch of elbow grease applied, you can get it to more like this. (YMMV/things are changing fast) "Please stand by... Running (5) subagents \- reading wikipedia across 12 languages \- finding references through pubmed \- searching bioarchive \- tracking down prepubs of leading scientists \- latest news articles Done. After 4m25s, 5 sub-agents have returned their search results. Summary and synthesis of 128 source documents is as follows... " [2] Which I'd suggest is a bit less likely to result in slop. The only reason you might not have heard of the latter approach is because it costs Real Money(tm) yet. But open weights models are catching up very fast, and there are multiple agentic harnesses that utilize them. Then you 'only' need a $4000 computer, which more people do actually have. And starting from non-existent a couple of years ago, local models are now getting cheaper every month; small ones (not yet good enough for this) do actually fit on phones. So you decide what that means for "AI gets better result than our articles" sincerely, Kim Bruning For rhetorical/email purposes the two examples are stylized and cleaned up; real world results tend to be messier and require interpretation, disclaimers apply, etc etc. [1] Qwen local with default/low prompting actually still looks a lot like this! [2] Loosely based on claude-code-in-docker runs a few months ago. On Thu, Apr 30, 2026 at 07:36:22AM +0200, Anders Wennersten via Wikimedia-l wrote: > Or we can react constructive and start think of Wikipedia in term of > > *articles with accesses still are up and content not reused by AI (yes those > exist, like (list of) opinion pollings for elections) > > *articles with content is fully reused by AI (I see no basic problem there) > > *articles where AI gets better result than our articles, (and here have to > think through how to react, change our behavior) > > Anders > > Den 2026-04-30 kl. 07:07, skrev Luis Villa via Wikimedia-l: > > The fundraising team would be the first to say that revenue is the > > means, not the end. They can tweak the numbers by running more/different > > banners, but that does not mean the project is healthy. > > > > Bigger picture point: at WikiCredCon, it was suggested that this year's > > strategic discussions would focus on last year's 6-8% decline. > > > > I would go a much different direction. After the 1st quarter, *our > > strategic process has to start by asking the hardest possible question: > > what if "reading an encyclopedia" is mostly over, like reading a print > > newspaper?* In other words, what if our long-term on-wiki readership > > graph looks like this one? > > https://www.pewresearch.org/journalism/fact-sheet/newspapers/ > > > > *So, for purposes of strategic discussion: assume that within 2-3 years > > our on-wiki readership will be at 5% of peak. Or even braver: assume > > that on-wiki readership goes to essentially zero, with only editors > > reading it. *I'm not saying this is the most likely outcome, just that > > approaching the question from that frame is much more likely to generate > > interesting and important ideas than trying to slowly tweak our decline. > > > > What then? > > > > "zero on-wiki readers" wouldn't mean our mission is over: there are many > > ways (including many new and fun and productive ways!) to collaborate on > > open knowledge, many partners we could co-create with, and many channels > > to distribute that knowledge. And despite the general downturn, at least > > some outlets adjacent to us are thriving right now: > > https://lu.is/2026/04/web-collaboration-five-graphs/ So I'm not saying > > "all is doomed". > > > > But approaching our mission *without* assuming the magnetic force of the > > wiki would force us to be creative--to start experimenting widely and > > boldly. I think we need to place 1,000 bets with 10,000 motivated > > volunteers, sooner rather than later. > > > > This is not going to be easy. Among other things: > > - lots of old habits to unlearn. (I admit my own first public experiment > > is on... en-wiki. Hopefully #2 will be on wikidata, at least! And I've > > finally bought a .wiki domain for #3...) > > - if those 10,000 volunteers have to do this entirely on their own time > > (and own GPUs) then we'll again be building a movement by and for the > > (relatively) wealthy. Hopefully we can find many funders, not just WMF, > > so that we have a diversity of approaches and risk-tolerances. > > - lots of those experiments are going to fail, and we're going to have > > to pick ourselves up and try again anyway. (Shades of Mako's paper: > > https://mako.cc/academic/hill-almost_wikipedia-DRAFT.pdf) > > - we're going to have to find new signs of success. It might be readers, > > but it might not, or it might be different kinds of readers, or it might > > more explicitly be GLAMs, or... we don't know yet. But the One Big > > Number of readership likely will never be The Thing again. > > > > Anyway, I'm mostly going to go back to experimenting (and hopefully > > sharing more about that experimentation). But had to get that off my > > chest. Hope it spurs others to plant some seeds and run some > > experiments! > > > > Trying to be Bold- > > Luis > > > > On Wed, Apr 29, 2026 at 6:30 PM Anders Wennersten via Wikimedia-l > > <[email protected]> wrote: > > > > Facts states (worldwide figures): > > > > *Donation is up > > > > *Number of new pages is up [1] > > > > *number of edits från users is up [2} > > > > That is not figures of a crises > > > > The 20% decrease represent a different way of accessing info. we > > have to learn how to adjust to these changes but is is NOT a > > general crises, > > > > Anders > > > > [1] > > https://stats.wikimedia.org/#/all-wikipedia-projects/contributing/new-pages/normal|bar|2-year|page_type~content|monthly > > <https://stats.wikimedia.org/#/all-wikipedia-projects/contributing/new-pages/normal%7Cbar%7C2-year%7Cpage_type~content%7Cmonthly> > > > > [2] > > https://stats.wikimedia.org/#/all-wikipedia-projects/contributing/edits/normal|bar|2-year|editor_type~user+(page_type)~content|monthly > > <https://stats.wikimedia.org/#/all-wikipedia-projects/contributing/edits/normal%7Cbar%7C2-year%7Ceditor_type~user+(page_type)~content%7Cmonthly> > > > > Den 2026-04-28 kl. 17:31, skrev Christophe Henner via Wikimedia-l: > > > Hi, > > > > > > 3 months later, a new statistic: mobile page views dropped by 19% > > > in one year. > > > > > > We are back to the mobile traffic levels of 2018. > > > https://stats.wikimedia.org/#/all-projects/reading/total-page-views/normal|bar|2016-01-01~2026-04-28|(access)~mobile-app*mobile-web+agent~user|monthly > > > <https://stats.wikimedia.org/#/all-projects/reading/total-page-views/normal%7Cbar%7C2016-01-01~2026-04-28%7C(access)~mobile-app*mobile-web+agent~user%7Cmonthly> > > > > > > At this rate, our mobile traffic in 2026 will be at the levels of > > > pre-2013. 2013... > > > > > > In the meantime, Ericsson is reporting a global mobile usage > > > growth of over 20% > > > https://www.ericsson.com/en/reports-and-papers/mobility-report/dataforecasts/mobile-traffic-update > > > > > > When I published my essay, I mostly received gentle to > > > aggressive push back. Back then I was displaying a 8% drop. Now a > > > 20% year over year. > > > > > > Why did I pick mobile traffic? Because historically bots are less > > > using desktop browser UAs which makes the data a bit more > > > reliable for trends. > > > > > > 3 months ago I said we had 24 months ahead of us before > > > irrelevancy. Well at this rate, in two years we will have lost > > > half of our traffic. And the sad part is that those downhill > > > trends tend not to slow down. > > > > > > And what now? Well being very blunt, there's only one body in our > > > movement that can muster leadership, energy and resources to > > > tackle this topic head on, that's the board of trustees. > > > > > > And if 20% doesn't trigger fast and real actions, I have no idea > > > what would. > > > > > > PS: yes this email is written with a lot of emotion, but minus > > > 20% YoY should trigger that reaction in everyone. > > > > > > -- > > > Christophe > > > > > > > > > On Sat, 10 Jan 2026 at 00:40, Christophe Henner > > > <[email protected]> wrote: > > > > > > Hey everyone, > > > > > > I struggled with the object, the most honest one would be > > > "Last exit before irrelevance" but that would be a bit violent. > > > > > > So, Wikipedia turns 25 next week. I've been here for over > > > twenty of those years, including stints chairing Wikimedia > > > France and the Foundation Board. And honestly? I'm worried. > > > Scared to be honnest. > > > > > > Over the last years, I've been regularly crunching data and > > > sharing on different channels my worries. But in the last few > > > weeks I decided to make a much more structured "essay" of my > > > findings > > > > > > Since 2016, the internet nearly doubled in size. > > > > > > Our page views? Down. New editor sign-ups? Down 36%. > > > > > > The people keeping this thing running are working harder than > > > ever, but there are fewer of them every year. > > > > > > I wrote it all up: the numbers, what I think went wrong, what > > > I think we need to do about it. Fair warning: it's long, it's > > > opinionated, and some of it will probably make you mad. > > > > > > *Here it is: > > > *https://meta.wikimedia.org/wiki/User:Schiste/what-now > > > > > > I'm not trying to be doom and gloom for the sake of it. I > > > genuinely believe we have maybe two years to make some hard > > > calls about AI, about money, about who we're actually > > > serving. After that, the window closes and we become irrelevant. > > > > > > Could be wrong. Hope I am. But I'd rather we have this > > > argument now than wish we had later. Well I'd rather we had > > > this argument two or four years ago, but now we will make do. > > > > > > Read it, tell me where I'm off base. Let's argue and debate. > > > That's what talk pages are for, right? > > > > > > PS: Foundation board mailing list is bcc'ed, change cannot > > > happen without their commitment. And fast. > > > > > > -- > > > > > > Christophe > > > > > > > > > _______________________________________________ > > > Wikimedia-l mailing list [email protected], guidelines at:https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines andhttps://meta.wikimedia.org/wiki/Wikimedia-l > > > Public archives athttps://lists.wikimedia.org/hyperkitty/list/[email protected]/message/VPVIIGLP2GKBMWTQCJKHRAXSMGBQHXJ5/ > > > To unsubscribe send an email [email protected] > > _______________________________________________ > > Wikimedia-l mailing list -- [email protected], > > guidelines at: > > https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and > > https://meta.wikimedia.org/wiki/Wikimedia-l > > Public archives at > > https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/KMB76EUVKIGGHKR6EPFGYEIZYTY3UYHV/ > > To unsubscribe send an email to [email protected] > > > > > > _______________________________________________ > > Wikimedia-l mailing list [email protected], guidelines at:https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines andhttps://meta.wikimedia.org/wiki/Wikimedia-l > > Public archives athttps://lists.wikimedia.org/hyperkitty/list/[email protected]/message/CRRVYOBW46THRH2WZ7OVTEUVFJKBM6NT/ > > To unsubscribe send an email [email protected] > _______________________________________________ > Wikimedia-l mailing list -- [email protected], guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l > Public archives at https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/7ZKXIME2QY7HMPH54LSUTHUAVVGDYVCK/ > To unsubscribe send an email to [email protected] _______________________________________________ Wikimedia-l mailing list -- [email protected], guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/RXZR3X4RBOJXODOVCQX5PNXTQT32TJAM/ To unsubscribe send an email to [email protected]