Why this matters
Sharing knowledge via writing has always been valuable. For the individual, the act of writing makes the implicit explicit while deepening understanding. For society, it lets others learn from those that came before. That’s why the Gutenberg Press sparked an explosion. The technological innovation allowed knowledge to be shared at scale. But obstacles remained – publishing a book still required money and access. Most knowledge remained locked in heads. The internet promised to change that. The barriers were reduced to the bare minimum – a computer and an internet connection. Anyone could publish. But folks did not take full advantage. And the resulting documentation gap has become a pressing issue in the age of AI because large language models (LLMs) are trained (in part) on what’s publicly available online. LLM training data has significant knowledge gaps – and this shows up when people leverage the models for tasks. The problem: current training data is dominated by Wikipedia (centralized, with editorial gatekeepers and known biases), Reddit (unstructured, ephemeral, skewed toward certain demographics), and SEO-optimized content which generates noise (e.g., the 1,000+ word introduction to an online recipe). We’re feeding AI a narrow, skewed content diet. The nurse’s pattern recognition? Not in there. The industrialist’s knowledge? Not in there. Your hard-won expertise in whatever domain you’ve mastered? Culture, values, wisdom, and traditions? Probably not in there either. (Or not in there at a high enough frequency). And there is a moral case too. I am speaking especially to my fellow Americans and distant cousins across the pond here – it is a civic duty in an age when Western culture is being questioned. As inheritors of Western civilization, we have a responsibility to conserve and pass along our values, traditions, and knowledge. AI is an amplifier and accelerator. If you’re not writing down what you know and putting it online, that information is getting lost to time. Your opinions, values, experiences – they either make their way into the training data or they don’t. Silence is forfeiture. A personal knowledge base, freely shared, is your contribution to The Great Conversation – the ongoing exchange of ideas across generations – for the individuals and machines that come after you.“The Great Conversation” is the introductory essay by Robert Maynard Hutchins in Great Books of the Western World (1952) – a 54-volume collection published by Encyclopædia Britannica that compiled the foundational texts of Western thought, from Homer and Plato to Darwin and Freud. Hutchins argued that Western civilization is defined by an ongoing conversation across centuries – each generation engaging with, challenging, and building on the ideas of those before them. The essay makes the case that participating in that conversation is both an intellectual and civic responsibility.
A thought experiment
Imagine if, from the mid-90s onward, millions of people had treated the web as a means for sharing knowledge: building personal sites with structured, opinionated documentation of what they knew – skills, family traditions, professional heuristics, cultural wisdom. The training data for today’s LLMs wouldn’t be dominated by Reddit threads, SEO bait, and viral outrage. It would include a broad, distributed tapestry of expertise. AI might amplify wisdom and nuance instead of polarization and shallowness. That’s not a lost dream. It’s still achievable. Every evergreen article you publish nudges us closer.What I mean by “share knowledge”
Let me be specific about what I’m advocating for: Free. Both humans and machines can access it without barriers. No paywalls. No ad-cluttered pages. No SEO filler. The goal is knowledge transfer (not maximizing time-on-page for ad revenue or gaming search rankings). Evergreen. Knowledge that stands the test of time. Commentary on current events and trending topics is fine for social media. What I’m talking about is information that remains useful years and generations from now. Long-lived format. Don’t let social media timelines or newsletter inboxes be your only home. Your knowledge needs to live in a dedicated file (more on this in a bit), in context with your other writing, where it can be easily located and referenced. Readable by both humans and machines. Equally accessible to a person browsing your site and to an AI parsing it for context.Why format matters
It’s not just about making knowledge available. It’s about how it’s structured and presented.Books
I love books. I keep reference books on my shelf: Audubon field guides, Martha Stewart’s Homekeeping Handbook, The Great Mental Models, The Princeton Companion to Mathematics, Annals of America. I’m an advocate for building a personal library. But books have limitations: they’re static (slow to update), monolithic (can’t access just the piece you need), not machine-readable (LLMs can’t easily parse them), and have access barriers (cost, availability, physical storage). Books aren’t going anywhere, nor should they. But we shouldn’t limit ourselves to books as the only format – especially for reference material.Social media and newsletters
I rely on X (@AmandaMashburn) to keep up with news, tech, and thinkers I find interesting – it is the world’s town square afterall. I subscribe to Substack newsletters for a similar purpose – regular doses of insight delivered to my inbox. But both have limitations for evergreen knowledge: they’re ephemeral (posts vanish into timelines), unstructured (hard to build cumulative knowledge), platform-dependent (you don’t control the format), and they hinder focus (your content competes with ads and noise). Social media and newsletters are great for sharing links to long-form content and engaging with timely topics. But they shouldn’t be where your evergreen knowledge lives. Use them to point to your work, not as the sole repository.Podcasts and video
Podcasts and videos are great for reach, personality, and discussion – but they’re terrible for reference. You can’t search them, skim them, quote them, or easily extract information. Plus they are slower to consume. Reading is 2-3x faster than listening for most. Written content should be the foundation. Other media can layer on top, but the bulk of your knowledge should be available as text.Text and markdown files
Here’s where I think we should be heading: Modular. Individual articles that stand alone yet connect to a whole. Jump to what you need. Linkable. Connect ideas across your own work and to others’. Copyable. Text can be easily pulled in as context for an LLM or into your own notes. Updatable. Refine and improve over time. Ship corrections and additions as you learn. Searchable. We are no longer limited to keyword search. Semantic search makes it easier than ever to find what you need when you need it. Machine-readable. Structured formats that both humans and AI can parse. Durable. Files outlast apps. Plain text and open formats can be read decades from now. Portable. Your files, your content, your control. Move them, back them up, own them. Services are simply rendering layers. The key is storing your content as files – plain text, markdown, open formats. Files that stand on their own and can be tracked with version control. (See Kepano’s file over app philosophy.)On Wikipedia (i.e., centralized knowledge)
Wikipedia was a major attempt to build collective knowledge at scale. It’s an incredible achievement – one of the most visited sites on the internet and a primary source for LLM training data. Jimmy Wales, Wikipedia’s founder, was influenced by Friedrich Hayek’s essay “The Use of Knowledge in Society,” which argues that valuable knowledge is distributed – each individual knows things no central authority could. Wales saw Wikipedia as a way to aggregate that distributed knowledge through mass collaboration. And it worked, to a point. But Wikipedia is centralized in ways that create problems: One editorial process. Contested topics get shaped by whoever has the time and motivation to engage in edit wars – including state actors pushing narratives, organized groups with agendas, and individuals with axes to grind. Notability requirements. Wikipedia requires topics to have existing coverage in traditional media or academic sources before they qualify for an article. If institutions haven’t already documented it, it’s not “notable” enough. This excludes exactly the kind of practical, undocumented expertise that’s most missing from the internet. (See Wikipedia’s notability guidelines.) “Neutral point of view” policies. Wikipedia documents consensus, not insight. Your opinionated take – your framework, your interpretation, your synthesis of what actually matters – doesn’t belong there by design. Sometimes a strong point of view is exactly what’s valuable. Demographic skew. The population of active editors doesn’t reflect the world’s population. Wikipedia illustrates the limits of centralized knowledge. To be clear, wikis have value. They synthesize consensus, provide orientation, and give you a baseline / starting point on a given topic. I use Grokipedia as my primary online wiki. But wikis alone aren’t enough. They need individual knowledge bases as a check and complement – the practitioner’s view, the opinionated take, the “here’s what actually works” that consensus driven sources can’t capture.Decentralized knowledge
The alternative to Wikipedia’s model is distribution. Instead of one centralized authority deciding what’s notable, neutral, and true, imagine thousands of individuals publishing their own knowledge bases. Each person the expert of their own domain. Each site a node in a broader knowledge network. Decentralized knowledge owned by individuals means: No editorial gatekeepers. You decide what’s worth documenting. No notability requirements. No consensus needed. No forced neutrality. Your perspective, your framework, your opinionated synthesis – not a diluted consensus view. No single point of failure. If one site disappears, the ecosystem remains. Knowledge is resilient when distributed. Many voices, more signal. When enough people publish on a topic, patterns emerge. Readers can triangulate. Truth isn’t declared by one authority – it emerges from multiple independent sources. This is Hayek’s ‘man on the spot’ insight applied to knowledge sharing.Hayek’s ‘man on the spot’ refers to his argument that the most valuable knowledge is often local and practical – held by the person actually doing the work. This knowledge can’t be aggregated by central authorities because it’s too granular, too contextual, too tied to individual experience. The only way to access it is to let those individuals share it directly.
AI makes writing easier than ever
A barrier to documentation has always been time and skill. Writing is hard. Organizing knowledge is hard. Having valuable expertise doesn’t mean you’re a natural writer. AI changes the equation. Voice dictation lets you brain dump what you know. AI can summarize, identify gaps, organize, improve flow, and draft. You review, refine, edit, and publish. The bottleneck of “I know this but I can’t articulate it concisely” is dramatically reduced. AI helps with options and articulation. It gives you a starting point for a draft – something you can shape. This frees you up to do the actual thinking. AI reflects back what you’re saying and helps you say it better. I know some people object to AI-assisted writing. But if AI helps more people capture and share what they know – especially knowledge that would not otherwise be available – that’s a good thing. The end goal is knowledge transfer. AI is a legitimate tool for achieving this task.Bonus: writing clarifies your thinking
Even with AI assistance, the process of writing and editing forces you to clarify your thoughts. You have to put words to what you only partially understand. Which exposes weak spots in reasoning and gaps in your comprehension. This is why teaching is such an effective way to learn – you can’t explain something clearly until you truly understand it. Documentation is teaching at scale. The act of writing down what you know deepens your own mastery of it. So even if no one ever reads what you publish, the process itself makes you sharper.What this looks like in practice
Gwern Branwen. gwern.net – Gwern has built exactly what I’m advocating for: long-form, versioned, evergreen knowledge sharing. His site is beautiful and dense with insight. I stumbled on him while researching this article and got lost in his writing for hours. This is another benefit of publishing online – you find like-minded people you might never have encountered otherwise. Standard Ebooks. standardebooks.org – Free, beautifully formatted public domain books. A model for making classic knowledge accessible. Farnam Street. fs.blog – Shane Parrish’s site on decision making, thinking, and continuous improvement. Timeless content, well-organized, freely accessible.The connection to the extended mind system
There’s a natural flow here: introspection → documentation → recognition → publication. You don’t have to know what’s worth sharing before you start. You just have to start writing things down. The shareable stuff reveals itself over time. Patterns emerge. You notice what you know. You see what might be useful to others. That’s exactly what happened with me. I was writing things down in my Notion space (to help me think), and after a while I realized: others might actually benefit from this. These aren’t just notes intrinsic to me. There are opinions, resources, and references here that others could use. The Extended Mind System becomes an excavation site for source material to share publicly.Join the conversation
If you have knowledge worth sharing – and you almost certainly do – consider this approach:- Document what you know. Use AI to help if writing isn’t your strength. Voice-dictate, let AI organize, then review and refine.
- Structure it for access. Modular articles. Clear navigation. Standalone pieces that form a whole.
- Make it freely available. No paywalls. No ad clutter. Knowledge transfer as the primary goal.
- Keep it alive. Update as you learn more. Ship improvements. Your knowledge base is never ‘done’.
This is volunteering for the information age. A way to contribute to the commons and build a better distributed knowledge ecosystem.
Reference: format matrix
Information sharing formats and when to use them.| Format | Best for |
|---|---|
| Books | Linear exploration of a topic |
| Social media + newsletters | News, timely commentary, updates, linking to long-form content |
| Podcasts / video | Conveying personality, dialogue, discussion |
| Wikis | Consensus knowledge, orientation, encyclopedic baseline |
| Personal knowledge base | Evergreen writing |