Logo

Civilization Scale Memory

Cage

For the past 20 years, one website on the internet has done more to tether the world to a consensus reality than any other website. That site is Wikipedia. And, for those who are old enough to remember, when Wikipedia started it was widely seen as a joke. There was no way a site that let anyone edit could be at all reliable. But it found a way and became one of the most visited sites in the world.

Flash forward to 2025 and we're in a different era.

AI powered chat bots, which trained on Wikipedia, have gained hundreds of millions of users and are taking mindshare away from Wikipedia. Instead of trusting Wikipedia's consensus mechanism where you can see the full history of edits of any given page, people are putting their trust in Sam Altman's LLM. A black box that the user can not see.

Meanwhile, Wikipedia has been slipping. Beyond losing market share the foundation that runs the site has had scandal after scandal and been credibly accused of ideological capture. And, despite an endowment of over 100 million dollars, an organizational structure that has led to stagnation.

People are rightfully wondering how a late stage Wikipedia will evolve. But even more generally, there is a question about the format itself.When chatbots are this good and we can summon a bespoke answer for any question you have, what is the value of an internet encyclopedia?

The case for a wiki style encyclopedia

  • Truth and accuracy
    • The most sophisticated models currently still hallucinate and struggle to correctly cite sources.
  • A ledger of record
    • It's important to keep track of what people think is true and know when it changes.
  • People arguing in public about what is true > Black Box Consensus
  • The encyclopedia format is still good for discovery
    • It's a different way to browse than a chat interface that does a good job of pulling you to related content.
    • Wikipedia rabbit holes are interesting for a reason.
  • The encyclopedia format is still good as a challenge
    • To better organize taxonomies of human knowledge

Ways to evolve an internet encyclopedia

  • Knowledge Graph > Article
  • Design for SEO and LLM output
    • RAG DB
  • AI agents as researches
  • AI agents as moderators
  • Better incentives for experts to join
  • Expand the depth and breadth of all entries

Fork Wikipedia

Wikipedia is a brilliant website. There has never been a more interesting time to explore that mission with a different approach.

Sprint 1

Forking Wikipedia, and it's 7 million entires, is a huge project. Before we commit to a fork we want to make sure we have enough of a hook to justify the transition. We're hacking on experiments to explore UX and researching approaches. Our first one is live.

This is currently a nights and weekends project. To support, back the project here.