Semantic tree learning, how I read for four ventures at once
You can’t run four companies by knowing everything about each one. The math doesn’t work. The calendar doesn’t work. Your brain doesn’t work that way.
What works is knowing the right things deeply and having a framework for what to skim. Years ago I came across a Musk quote about treating knowledge as a semantic tree, trunk first, branches second, leaves last. It was the single most useful learning frame I had encountered, and it continues to be. This post is how I apply it across very different categories, and where it quietly fails if you’re not careful.
The idea, quickly
A semantic tree is a model of knowledge shaped like a tree.
The trunk is the fundamentals. The laws and mechanics of a field, the things that are true below all the specifics, the principles that every specific instance is an expression of.
The branches are the applications. How those fundamentals show up in different contexts, categories, use cases.
The leaves are the details. The names of companies, the dates of events, the specific tactics, the vendor comparisons, the news of the week.
The Musk observation is that most people collect leaves without ever growing a trunk. They read twenty articles about a topic and know nothing durable because nothing they’ve read attaches to a foundation. A leaf without a branch falls off.
Leaves are cheap. Trunks are expensive. The cheapness of leaves is why we spend so much time on them and learn so little.
The trunks of my four ventures
Running across multiple categories has forced me to be explicit about what I consider the trunk of each. I don’t have time to learn every leaf. I need to know that when a leaf shows up, I can place it on the right branch of the right trunk and know what it means.
Scentpression, the trunk is perception. Underneath every scent marketing decision is the neuroanatomy of the olfactory system, the behavior of volatile compounds in conditioned air, and the psychology of how humans report subjective experience. Everything else, which diffuser, which fragrance house, which property type, is a branch. Vendors change. The biology doesn’t.
Vanbert Aviation Group, the trunk is aviation economics. Underneath every charter decision is the unit economics of an aircraft, the regulatory constraints on crew time, and the capital structure of operators. Empty leg markets, jet cards, fractional programs, on demand brokerage, all branches of that trunk. A new platform launches. A new business model appears. You can evaluate either of them in ten minutes if you have the trunk.
Bahamas Outdoor Media Ltd, the trunk is attention. Underneath every media buy is how humans pay attention in physical space, how attention converts to memory, and how memory converts to action. Billboards, shelters, airport halls, digital OOH, branches. The leaves are placement lists and rate cards. A new format appears, you place it on the tree and know what it’s good for.
One Investment Group, the trunk is capital allocation. Underneath every investment decision is the cost of capital, the compounding of time, the asymmetry of downside versus upside, the optionality of position sizing. Hospitality deals, real estate, fintech, logistics, branches. Specific companies, specific deals, leaves.
Four trunks. Not four encyclopedias. The encyclopedia is what people think they need. It is not what they need.
How I actually build a trunk
The hardest part of semantic tree learning is identifying the trunk in a category you are new to. If you pick the wrong trunk, everything downstream rests on an unstable foundation.
Three tests I use.
One, the persistence test. If a principle has been true for fifty years across different technologies and different market conditions, it is probably trunk. If it has been true for eighteen months, it is probably a branch or a leaf. The neuroscience of olfactory memory is trunk. The current market leader in HVAC diffusion is a leaf.
Two, the translatability test. If a principle applies to multiple apparently different categories once you understand it, it is probably trunk. Cost of capital is trunk, it applies to hospitality, real estate, fintech, and every other category I touch. A specific valuation ratio for hotels is a branch.
Three, the first principles test. If I can derive a principle from something more fundamental, it isn’t the trunk, the thing I’m deriving it from is. I keep moving down until I hit something irreducible. That’s the trunk.
Once you find the trunk, you read it slowly. You read it multiple times. You take notes by hand. You come back to it a quarter later and re read it. The trunk takes real time and there is no shortcut.
The branches you can read faster. The leaves you can scan.
Where most operators get this wrong
Two common failure modes I’ve watched.
Collecting leaves forever. An operator spends years reading news in a category. Trade publications, LinkedIn posts, quarterly earnings summaries. They know every current leaf. They cannot tell you why any of it matters. When the category shifts, they are lost because they have no trunk to orient on.
Mistaking a branch for the trunk. An operator specializes early, say, in a specific vendor’s platform, or a specific regulatory regime, and treats that specialization as foundational. When the platform gets disrupted or the regulation changes, the specialization becomes a liability. They have to regrow the tree from scratch because the thing they were rooted in was never the trunk.
Both failure modes are common because the trunk is the slow, boring part. Leaves produce the feeling of productivity. Branches produce the feeling of expertise. Only the trunk produces durable understanding, and it does so quietly over a long time.
The operator’s version, specifically
For a founder or specialist, the semantic tree can go very deep in one area. You can know the trunk and many branches with real fluency in a single category.
For an operator running across categories, the trees stay shallower but you have more of them. My tree for aviation economics is not as deep as a career aviation analyst’s tree. But it is a real tree, rooted in real trunks, and it allows me to make decent decisions at the operator level in a category that isn’t my first one.
The win is not being the world expert in four categories. That is impossible. The win is being a competent operator in each, with enough of a trunk to avoid catastrophic errors and enough humility to hire the specialists who know the branches better than you do.
Where this meets first principles
Semantic tree learning and first principles thinking are two halves of the same practice. First principles is how you interrogate a trunk, making sure it’s actually true and not merely conventional. Semantic tree is how you architect what you learn once you’ve confirmed the trunk is real.
I’ll write the synthesis post separately because it deserves its own essay. For now: if you’re only using one of the two, you’re doing half the work. First principles thinking without the architecture spirals into perpetual questioning. Semantic tree without interrogation memorizes a conventional foundation that may not survive scrutiny.
Together, they’re how I try to learn anything I need to learn at operator speed, across categories that don’t look like each other.
Further reading: First principles thinking · Contrarian thinking from the Caribbean