The First Principles AI Learning Method

Most students begin by opening Chapter 1 and reading forward. The problem is structural: textbooks are organised for completeness, not for understanding. Every chapter is a detail. By the time you reach Chapter 8, you have often forgotten what Chapter 2 said, because nothing told your brain why it mattered.

The First Principles approach inverts this. Before learning any detail, you ask: what are the irreducible truths this subject is built on? In physics, that might be Newton's three laws of motion and the conservation of energy. In economics, it might be scarcity, incentives, and marginal thinking. In biology, it might be evolution by natural selection, cell theory, and homeostasis.

Educational psychologist David Ausubel called these "advance organizers" — brief, high-level overviews given before studying new material. Across dozens of experiments published throughout the 1960s and 1970s, students who received an advance organizer consistently outperformed those who began with the details directly. The effect was largest when the material was most unfamiliar. The organizer gives the brain a skeleton to hang information on. Without it, facts arrive with nowhere to go.

Elon Musk has described his own learning method in similar terms — "boiling things down to the most fundamental truths and then reasoning up from there," tracing the idea back to Aristotle's first principles in Posterior Analytics. The principle scales from physics to any subject you need to learn quickly.

Once you have the foundations, the next step is to see how everything else branches from them. Elon Musk has described this as building a "semantic tree" — ensuring the trunk and main branches are solid before worrying about the leaves. The analogy is cognitively precise.

John Sweller's cognitive load theory (1988) explains why the order matters: working memory can only hold roughly four independent pieces of information simultaneously. When you study without a mental map, each new concept has to compete for that limited space. When you already have a tree structure, a new concept does not need to be stored as an isolated fact — it slots onto a branch you already have, requiring far less cognitive effort.

Research on concept mapping by Joseph Novak and D. Bob Gowin (Learning How to Learn, 1984) found that students who created hierarchical visual representations of a subject before studying the details scored significantly higher on retention and transfer tests than those who studied the same material linearly. The key was not the drawing itself, but the act of deciding what was a trunk, what was a branch, and what was a leaf — the process forced decisions about importance and relationship that linear reading does not.

Use this prompt to build a complete structural map of any subject. It will not replace studying the details — but it will make everything you study afterwards feel like it fits together rather than arriving in disconnected pieces.

Richard Feynman — Nobel laureate in physics and arguably the most effective science teacher of the 20th century — had a simple test for whether he truly understood something: he would try to explain it in plain language to someone with no background in the subject. If he could not, he would return and study until he could. The method has become known as the Feynman Technique.

The reason it works is well-documented. Arthur Glenberg and William Epstein (1985) identified what they called the "illusion of knowing": when you read something, your brain creates a feeling of familiarity that can be easily mistaken for understanding. You think you know it. When asked to explain it from scratch, without notes, the gaps become immediately visible — not because you have forgotten it, but because you never fully understood it in the first place.

Retrieval practice research by Henry Roediger and Jeffrey Karpicke (2006), published in Psychological Science, confirmed that students who were tested on material — even when initial test performance was imperfect — retained significantly more one week later than students who spent the same time re-reading. The act of attempting retrieval, including the moments of failure, is itself a learning event that re-reading cannot replicate.

The prompt below uses AI as your Feynman partner. Write your explanation in plain language without looking at your notes. Then let the AI find exactly where your understanding has holes — and ask follow-up questions to make you locate the gaps yourself rather than simply telling you the answers.

The final step is the most underused. When you connect a new concept to something you already understand from a completely different domain, two things happen simultaneously: the new concept becomes anchored to an existing strong memory, and the comparison often reveals structural patterns you had not noticed. Physics starts to resemble economics. History starts to resemble biology. This is not coincidence — it is the same underlying structures appearing in different domains.

Dedre Gentner's structure-mapping theory of analogical reasoning (Cognitive Science, 1983) explains the mechanism: the brain does not store just facts — it stores relationships between facts. When you recognise that two different domains share the same relational structure, understanding one deepens your understanding of both. The analogy works not because the surface details match, but because the way the parts relate to each other is identical.

A related technique, elaborative interrogation, was systematised by Michael Pressley and colleagues (Journal of Educational Psychology, 1987): the habit of asking "why is this true?" and "how does this relate to what I already know?" during learning — not after — produces 30–60% better retention than passive re-reading in controlled studies. The mechanism is the same: connecting new information to prior knowledge creates multiple retrieval pathways into the same memory, making it harder to forget.

The important discipline in using analogies is identifying where they break down. An analogy that is pushed too far becomes a misconception. The prompt below is built around this — it asks the AI to be explicit about the limits of the comparison, which stops you from misapplying it.