Learning By Writing — Holden Karnofsky

Holden Karnofsky describes his method for going from "never heard of this" to "detailed opinions with retained reasoning." The approach inverts the usual order: organize learning around writing, with reading always in service of forming and defending a bottom-line view.

The basic cycle is:

1. Pick a topic (claims that are important if true and might be true; pay attention to smart, unconventional voices).
2. Read and/or discuss the prominent 1-3 pieces on each side (a bit).
3. Explain and defend your current, incredibly premature hypothesis in writing (bold and aggressive).
4. Find and list weaknesses in your case.
5. Pick a high-leverage subquestion and do more targeted reading/discussion.
6. Revise your claim or switch sides.
7. Repeat steps 3-6 many times.
8. Get feedback on a draft and keep repeating.

Step 3 is the unusual and high-leverage part: you form and articulate a hypothesis long before you are "qualified." This forces reading to stay focused on the goal of a bottom-line view rather than open-ended information gathering.

Why Write Early and Often

Writing the hypothesis down early reveals which version of the claim you actually care about and which sub-questions are worth deep dives. Debates are multi-dimensional; without an explicit working hypothesis you risk disappearing into exhaustive but low-leverage reading on every dimension.

The process deliberately produces many premature, wrong hypotheses that then get stress-tested. "This generally fails almost immediately" is the point.

Devil's Advocate and Revision

After writing, play devil's advocate against yourself: seek the strongest contradicting arguments and note where your current case is thin because you have not looked comprehensively.

Then pick the sub-question most likely to change the headline claim and investigate narrowly.

Revision includes actively trying the opposite side: "I often try radically flipping around my hypothesis, even if I haven’t actually been convinced." Getting the best arguments for both sides is more productive than trying to be even-handed from the beginning.

Repeat the write–critique–targeted-read–revise loop until the draft feels like the best case you can make without vastly more work.

Example

Karnofsky uses the running example "Human history is a story of life getting gradually, consistently better." Initial hypothesis after light reading, then flip to "hunter-gatherers had important things right that modernity lost," then detailed sub-investigations on violence and gender relations that produced separate pieces, then return to the original headline with more nuance.

The final outputs are opinionated, evidence-grounded, and far more retainable than passive reading would have produced.

Relation to Strategy

This method is a concrete implementation of several of the non-automatic strategic heuristics (see humans-are-not-automatically-strategic): defining what success on the investigation looks like, gathering targeted info, systematic testing of conjectures (your successive hypotheses), focusing effort on what can actually change the view, and making the goal (a bottom-line opinion) explicit and coherent.

See also its appearance in the generalist reading list as a recommended piece on strategy and information-gathering: a-reading-list-for-generalists-dylan-bowman.

Cross references: match-quality polymath a-reading-list-for-generalists-dylan-bowman.