Three Websites, One Rule: Why Google Is Disqualified

The rule is simple: Google is not allowed. Google contains everything, so choosing it is not a decision. It is a refusal to decide. Removing it forces a better question: what three tools, each doing a different job, would together cover the path from raw information to durable knowledge?

My answer is Polymarket, Claude, and FMHY. They are not three versions of the same thing. They sit at three different stages of how a person comes to know something.

There is a method to the constraint. Default tools decide for us quietly: you type, you click the first plausible result, you move on feeling informed. Removing the default breaks that loop and forces the harder question underneath every search — what am I actually trying to do here? Am I orienting, gathering, or understanding? The three tools I kept each answer exactly one of those questions, and none of them lets me pretend I have done all three.

The pipeline

Information is not knowledge. Information is a signal, and knowledge is a signal that has been located, tested, and integrated into how you reason. Most of what we call "research" fails at one of three points: we do not know what is worth attending to, we cannot find good material, or we cannot turn material into understanding. Each site addresses one of these failures.

It helps to be precise about the stages, because each one fails differently. Information becomes evidence when it carries provenance — a source, a method, a reason to trust it over the alternatives. Evidence becomes understanding when you can argue from it, not just repeat it: when the model in your head makes predictions that survive contact with disagreement. And understanding becomes durable knowledge only when it holds across contexts, when you can rebuild the reasoning months later instead of merely remembering the conclusion. The pipeline leaks at every joint, which is why a single tool never covers it.

Polymarket: information with a price on it

I do not use Polymarket to bet. I use it because it changes the incentive structure of information.

Most commentary is free to produce and costless to get wrong. A pundit who predicts an outcome loses nothing when the prediction fails. A prediction market removes that immunity, because being wrong costs money. Hayek argued in 1945 that prices aggregate dispersed knowledge that no single person holds, and a prediction market applies that idea to questions about the future. The academic literature, including Wolfers and Zitzewitz's work on prediction markets, generally finds that market probabilities are informative and often well calibrated compared with polls and expert panels.

For learning, the value is not that the odds are always right. It is that they give you a quantified baseline to disagree with. When I read a confident analysis and then check the market, the gap between the two is a question worth investigating.

Over time this becomes a habit of mind rather than a website habit. Putting numbers on beliefs — even rough ones, even privately — makes overconfidence visible to yourself. Markets also teach base rates without lecturing: most long shots fail, most confident narratives dissolve, and the world is wider than any single story about it. Calibration is not a talent; it is a practice of keeping score, and a market keeps score whether you feel like it or not.

The caveats matter. Thin markets are noisy. Long-shot outcomes tend to be overpriced. Wealthy participants can move prices for reasons unrelated to belief. And every market depends on its resolution criteria, so you must read the exact wording of what counts as a "yes." Treat a probability as a prior, not a verdict. Legal access also varies by jurisdiction, so check what applies to you.

Claude: from signal to understanding

A probability tells you what a crowd believes. It does not tell you why, or what follows. This is the stage where a reasoning partner is useful. I use Claude to stress-test an argument, translate between disciplines, build a model of a problem, and ask the question I did not know to ask.

The danger is well known and I have written about it elsewhere: the ELIZA effect, where fluency is mistaken for reliability. A language model can be articulate and wrong in the same sentence. So the discipline is to use it as an interlocutor and not as an oracle. Ask it to argue against your position. Ask it what evidence would change its answer. Verify anything load-bearing against a primary source. Used this way, it compresses the distance between confusion and a testable understanding. Used carelessly, it produces confidence without contact with reality.

What I value most is the way a good session surfaces the assumptions I did not know I was making. Left alone, my thinking follows the groove of whatever I already believe; the model, prompted adversarially, keeps proposing the second and third explanations I would not have reached for. Steelman the opposition. List the competing hypotheses and what each one predicts. Name the experiment — or the document, or the dataset — that would settle it. The point was never to outsource judgment. It is to multiply the judgments available and then choose among them with open eyes, checking the load-bearing ones against sources that exist outside the conversation.

FMHY: the map of the territory

The third choice is the least obvious. FMHY is a community-curated index of free resources, covering tools, learning material, open-source software, reference sites, and much else. Its function is discovery. Search engines are good at answering questions you can already phrase, but they are poor at showing you what exists. A curated index reveals options you would not have thought to query.

This matters because access shapes what a person can learn. A student with a good map of free tools and materials competes with someone who has a large budget.

There is also a quieter benefit that search engines rarely provide: serendipity with structure. Browsing a curated index, you collide with adjacent fields — a tool built for one purpose that solves your problem, a reference work from a discipline you have never entered. Queries return what you asked for; maps show you what is nearby. For a learner without institutional access, that difference compounds. Each free resource found is not just one answer but a new direction the next question can come from.

Two honest caveats apply. First, FMHY is community-maintained, so quality varies and each link deserves scrutiny. Second, parts of it point to material of doubtful legality. I am recommending it as a discovery layer for legitimate learning and tooling, and I would encourage using the legal sections and paying for what you can afford to support the people who make things.

Why these three fit together

Each covers a weakness in the others.

• Polymarket tells you what matters and how uncertain it is, but it cannot explain itself.
• Claude explains and reasons, but it can be confidently wrong and has no stake in being right.
• FMHY shows you where the resources are, but it does not tell you what to make of them.

They form a loop. The market surfaces a question. The index gives you the material. The model helps you reason through it. Then you return to the market and check whether your understanding would have priced the outcome better than the crowd did.

Notice what this loop is really doing: it separates the stages that collapse into mush inside a single search box. Orientation first — what is uncertain, and by how much. Then gathering — the actual material, with provenance. Then reasoning — slowly, adversarially, in your own words. Most research fails because these stages are skipped or merged; the workflow works because each tool refuses to do the other tools' jobs. That refusal is the feature.

That last step is the real reason I chose Polymarket. Calibration, matching your confidence to the evidence, is the discipline underneath honesty with yourself. A prediction market is one of the few places where that discipline is scored.

Google would have given me everything, and that is why it fails as a choice. Everything, without a structure for using it, is just noise. Access was the bottleneck of a previous century; it is not ours. Our bottleneck is what we do after access — how we orient toward uncertainty, gather with provenance, and reason toward something that survives disagreement. The three sites I picked are a structure: signal, sense, and substrate.