Where good information actually comes from

Information quality falls as popularity and distribution efficiency rise.

Not because people are stupid. Because incentives degrade information as it scales.

A source that reaches ten people can afford to be precise, uncomfortable, and slow. A source that reaches ten million must be fast, frictionless, and emotionally resonant. Those constraints are not compatible with accuracy. Accuracy takes time, requires friction, and often feels bad.

This is not a new observation. But most people have never mapped it. They know that social media is noisy and books are valuable, but they have not thought through the mechanism or ranked their sources against it.

The useful concept is this: what does a source give back per unit of attention you spend on it? Some sources return compound insight. Others return the feeling of being informed without the substance. Both can feel identical in the moment.

I have found it useful to rank every source I use against this question. What follows is that ranking — in three layers, from individual to social to offline. It is not a universal prescription. It is how I think about where my attention should go.

Part 1: Solitary sources

These are the sources you engage with alone. No algorithm decides what you see next. No one else is watching.

Direct experience and building sits highest. The source is reality. Reality has no agenda. It only rewards or punishes. You write the code and it works or it does not. You launch the campaign and it converts or it does not. You build the habit and it compounds or it does not. The feedback loop is short and the data is real.

The cost of direct experience is that it is slow and expensive. You cannot experience everything. But the knowledge that comes from it is also the hardest to replicate or fake.

First-principles artifacts sit second — your own notes, post-mortems, decision logs, dashboards, processed observations. The source is you. The incentive is truth. Nobody else is reading it. There is no performance in it.

Most people underinvest here. They consume far more than they process. The gap between what you have read and what you have actually thought through is where most of the knowledge gets lost.

Books sit third. Good ones, not trending ones. Ten to twenty years of synthesis, no algorithm pressure, no engagement incentive. A book that has been in print for twenty years has survived a long filter. The author had time to think, an editor with standards, and a publisher who required coherence. That is a higher bar than almost anything published online.

The caveat is that books are slow to update. A book on machine learning from 2018 is mostly obsolete. The advantage holds for domains where the fundamental questions do not change quickly: psychology, history, business strategy, philosophy, anything about how humans work.

Courses built around execution sit fourth. Not video lectures with a completion certificate. Courses that require you to produce something — an assignment, a project, a real output with feedback. Most courses fail because they ship content instead of constraints. The constraint is the mechanism. It is what forces you to close the gap between knowing and doing.

Papers, technical docs, and manuals sit fifth. Narrow, accurate, high effort to parse. The upside is that they tend to be honest about what they do not know. The limitation is that most of the knowledge in them is inaccessible to a general reader without significant background.

Part 2: Social sources

These are the sources that involve other people — in your feed, in your inbox, in your community.

Newsletters and long-form conversations from genuine experts sit at the top of this layer. Not newsletters from people who write about growth hacking to an audience of growth hackers. Newsletters from practitioners who have done the thing and are reporting back. The difference is in whether the author has skin in the game outside the content itself.

Paid communities that require participation come next. Communities that only work if you contribute, extract, and ignore most threads. The operative phrase is "only work if." Most online communities are passive. The knowledge compounds only in the ones that demand output from members.

YouTube sits in the middle. Useful for discovery. Useful for "how does this work at a surface level." Poor for mastery. The format rewards demonstration over depth. You can learn the shape of a skill from YouTube. You cannot acquire the skill there.

X and LinkedIn are good for trends, signals, and finding people. They are poor for thinking. The format selects for provocation and brevity. Positions that survive there are the ones that can be stated in a sentence. Most valuable ideas cannot.

News is not for learning. It is for coordination and emotional priming. News selects for novelty and catastrophe. It systematically overrepresents rare events and underrepresents slow-moving trends. Reading the news produces a distorted model of the world that feels current but is not accurate.

Short-form video sits at the bottom of this layer. Zero declared intent, no retention mechanism, no feedback loop. You can scroll for two hours and be unable to name a single specific thing you learned. That is not accidental. The format is designed to continue consumption, not to transfer knowledge.

Part 3: Offline sources

This is the layer most writing on information quality skips. It is also the highest-trust layer.

Offline sources are defined by friction, presence, and consequence. Information that costs you something to receive tends to be more true. Information that you can receive with no friction, no presence, and no consequence tends to be less true.

Television and radio sit near the bottom of this layer — higher than short-form social, but still one-way and authority-biased. You receive what is broadcast. There is no feedback loop. Your reaction does not update the source.

Print journalism and books sit higher. Slower, more considered, subject to editorial process and legal consequence. Not immune to error, but operating under a set of constraints that social media does not.

In-person classes, lectures, and libraries sit higher still. You are physically present. You can ask questions. The teacher can see whether you understand. The feedback loop is tighter and the cost of your attention is higher.

At the top sit the sources with the most friction: apprenticeship, physical practice, deep one-to-one conversation, and direct observation of systems working and failing.

Apprenticeship is slow, expensive, and rare. It is also the most effective transfer mechanism for complex skill. The reason is that the master can see what you are doing wrong. They can correct in real time. They can tell you the things that cannot be written down because they only become visible in the act of doing.

Physical practice encodes knowledge in the body. Craft, sport, instrument, trade. Embodied knowledge survives changes in context that propositional knowledge does not. You can forget how to write about swimming. You cannot forget how to swim.

Deep one-to-one conversation with a practitioner is the fastest way to compress years of trial and error. One afternoon can cover what months of solo work misses. The condition is that both parties have to be willing to go beyond the performance layer.

The offline law

The more a piece of information costs you in body, time, and risk, the more true it tends to be.

Crafts beat content. Practice beats theory. Experience beats explanation.

This is not a romantic preference for difficulty. It is a description of a filter. Hard-to-access information has survived a rigorous test. The fact that it was hard to get is itself a signal.

Easy-to-access information has survived no such test. It may still be true. But you have no structural reason to trust it beyond the source's credibility, which you have to evaluate separately.

The five questions

Before trusting a source, I run through these. They work across every type of information.

1. Who benefits if I believe this? If the answer is "the person telling me this," that is not a disqualifier. But it is a flag. The stronger the financial or status incentive to make you believe something, the more scepticism is warranted.

2. Is there a feedback loop to reality? Can the source be wrong in a way that they would actually notice? If I believe this and act on it, can I find out whether I was right? Sources without feedback loops cannot update. They tend to drift toward whatever is comfortable or profitable to believe.

3. Is this built for truth or for attention? These are not always in conflict. But they are often in conflict. The structural incentives of mass-distribution platforms push toward attention. Ask which incentive is stronger.

4. Does it cost me anything to consume or apply? Zero-cost information is worth exactly what you paid for it on average. Not always. But the absence of friction is a warning sign, not a feature.

5. Will this change a decision I make this week? If not, it is probably entertainment. That is fine. But it should be labelled as such, not as learning.

The honest observation

Most people spend 70 to 90 percent of their attention in the bottom two layers of each tier. Social media, news, short-form video. That is where the defaults go. That is where the recommendation systems push you.

The result is that people feel informed without being informed. Confidence outpaces competence. The model of the world that accumulates from low-quality sources is not just incomplete. It is actively wrong in ways that are hard to detect because you are not running the information against anything real.

Knowledge does not compound automatically. It compounds when you build on a reliable foundation, correct errors through feedback, and apply what you learn to something with real stakes.

My actual intake target

This is what I aim for, not what I always achieve.

Roughly half my learning time is building and testing — projects, experiments, real work with real feedback. When something does not work, I learn more in that hour than in a week of reading about the topic.

A fifth goes to books and courses with real constraints. Slow, deliberate, often uncomfortable. I try to read at least one non-adjacent book every few months. Something that makes me a beginner again.

A fifth goes to curated sources — newsletters from practitioners, expert conversations, communities where the standard is high. I am ruthless about unsubscribing from anything that has started to feel like noise.

The remaining tenth is algorithmic and social. Mostly for scanning signals, finding people, and staying aware of what is moving. I treat it as discovery, not learning.

That mix is a target, not a record. Most weeks I do not hit it. But having the target makes it easier to notice when I have drifted into a feed at the cost of the other 90 percent.

The gap between what you consume and what you actually know is where the work is.