The world of human knowledge is split into two forms of data: structured and unstructured.
Programmers love structured data. It consists of keys and values, neat tables, and strict schemas. If you have five thousand books represented as structured rows in a database, asking a question like "What is the primary theme across all five thousand books?" is computationally trivial.
The problem is that real human experience doesn't arrive as a database. It arrives as unstructured prose. It comes in five thousand separate, messy volumes discussing love, heartbreak, physics, or finance.
Historically, if you wanted to query unstructured data, you had to perform an expensive pre-processing step: a human being had to read every page and take notes.
How Humans Learn
How does a human brain actually process new information?
It rarely starts from first principles. Instead, the brain learns through analogy.
Suppose a founder pitches you a new startup called "Chicken Magic." If they try to explain the raw mechanics—kitchen partner integration, fleet logistics, dynamic surge pricing—your mind has to do a lot of heavy lifting.
But if they say, "It’s Uber, but for fried chicken," your brain instantly understands. You already have a mental model for Uber, and you have a mental model for fried chicken. By linking those two existing nodes, the pitch creates a direct, high-bandwidth pathway to comprehension.
This is not just a trick for elevator pitches; it is the fundamental mechanics of human thought. When we encounter a novel concept, we immediately ask: What is this similar to? What known idea does this touch?
If a novel talks about "the one that got away," your brain doesn't just parse the seven literal words. It maps that phrase to an entire constellation of adjacent concepts: nostalgia, regret, romantic exaggeration, and human psychology.
How Machines Learn
Vector embeddings in modern language models work in the exact same geometric way.
Instead of matching literal keywords, an embedding projects words and concepts into a high-dimensional mathematical space. In that space, "the one that got away" sits right next to vectors for nostalgia and regret.
When you search across thousands of unstructured books using vector embeddings, the machine isn't reading word-by-word. It is performing a mathematical version of human analogy—retrieving concepts based on semantic proximity.
When machine embeddings and human analogical reasoning align, the way we consume knowledge changes completely.
The On-Demand Custom Book
Until recently, if you wanted to learn a subject, you had to read a book written by someone else for a generic audience. The author did the hard work of organizing their thoughts into chapters, sections, and paragraphs, but they had to pick a single reading level and narrative arc for all readers.