Skip to Main Content
Talk Beginner CC BY-SA 4.0

Let's Talk to Wiki

Proposal status is Rejected
Vysakh Premkumar
Vysakh Premkumar
Session Description

AI is cool, but I want the knowledge out there in public, not gated behind some mega-corp's servers. AI has already killed off a string of platforms by just answering the question directly instead of sending people to the source: Stack Overflow's traffic has collapsed since coders started turning to ChatGPT instead of waiting on a human answer, and Chegg cut 45% of its workforce, blaming the "new realities" of AI for its falling revenue.

It's essential we revisit Wikipedia and how it's managing this same AI crisis. Since January 2024, Wikimedia's bandwidth usage is up 50% not from human readers, but from AI crawlers scraping content to train models. Bots now account for at least 65% of the traffic hitting Wikimedia's most expensive datacenter infrastructure, despite being only ~35% of page views, and on top of that, human traffic is falling too, as AI tools answer questions using Wikipedia's content without sending readers back to the article. So the Foundation pays more to feed the bots while losing the human readership that funds it through donations, a genuine structural crisis for the largest open knowledge base on the internet.

MediaWiki, paired with MCP, is an actual viable answer to this: a way to make our archive AI-accessible without letting the knowledge get quietly strip-mined into someone else's model. We get to store the information ourselves, on our own infrastructure, under our own rules.

Key Takeaways
  • Keeping the source open: even if AI decides to hit the kill switch on a platform tomorrow, the knowledge should still be ours to access. Self-hosting the archive means we're never one corporate decision away from losing it.

  • A modern way of interacting with MediaWiki: using MCP and natural language as the interface, while Templates, SMW, and Cargo tables do the real work underneath, building richer and richer structured data instead of free-text drift.

  • The Direct Action Model: students get real access to the infrastructure itself, not a walled-off chatbot in front of it. That access is what lets them actually hold onto the narrative, instead of just consuming whatever version an AI hands back to them.

References

Session Categories

Introducing a FOSS project or a new version of a popular project
Tutorial about using a FOSS project
Community
Knowledge Commons (Open Hardware, Open Science, Open Data etc.)
Talk License: CC BY-SA 4.0

Which track are you applying for?

Main track

Speakers

Vysakh Premkumar
Vysakh Premkumar DevOps Engineer @ Open Healthcare Network | Open Healthcare Network

Vysakh Premkumar is a 2026 National Institute of Technology Graduate,
Who has 2 years of Industry Experience as a DevOps Engineer for the
10BedICU project, A volunteer for Open Healthcare Network and an active
FOSS Evangelist. He has been a speaker to conferences including but not
limited to the Open Suse Asia Summit 2024 Tokyo, COSCUP 2025 Taipei,
India FOSS 2024 Bangalore, Kochi Foss, and FOSSMeet at Calicut. He runs
and Maintains the wiki.fosscell.org the student owned collective
knowledge sharing platform using mediawiki. He finds interest in
building and deploying software that tries to solve the issues the
community he is part of generally face.