For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 780 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for the full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: the note below. ↓
A blog post titled "Stealing Is a Skill" made the rounds with a claim that sounds right until you press on it: that recreating someone's creation teaches you their craft, because in the act of rebuilding you encounter every tradeoff and imperfection they encountered. The comment thread underneath it did the real work, because it split cleanly down the middle, and the split exposes a distinction the post had blurred. There are two completely different acts wearing the same word, and only one of them teaches you anything.
The objection that lands hardest is that you do not learn by copying the output. You learn the destination, not the road. Reproduce the Mona Lisa stroke for stroke and you acquire none of the painter's relationship with the sitter, none of the cramped afternoons, none of the decisions to leave a thing unresolved. Worse, in software the copy actively misleads: people lift a pattern from a popular system without asking what the pattern was for, and propagate a solution to a problem they do not have, because the pattern's existence and popularity got mistaken for its appropriateness. Copying the surface does not transmit the reasoning that produced the surface. It transmits the surface, and the reasoning quietly does not come along.
But one commenter described the version that works, and it is not copying at all. Raymond Chandler taught himself to write a novelette by taking a story by Erle Stanley Gardner, writing a detailed synopsis of it, and then rebuilding the full novelette from the synopsis alone — never looking back at the original until his version was done. Then he compared the two and rewrote, and compared again, iterating until he understood the specific tricks Gardner had used to make a scene hold. The original was not the thing he copied. The original was the answer key. What he produced was his own reconstruction from a lossy compression, and the gap between the two was the entire lesson.
That gap is the mechanism, and it generalizes far past prose. When you compress a work down to a short description and then regenerate the work from the description, everything you regenerate correctly is something you already understood well enough to derive. Everything you get wrong is something the compression dropped that you could not put back, which means it was holding the structure up in a way you had not noticed. The residual after reconstruction is not noise. It is a precise map of your own misunderstanding, and it points at exactly where the next hour of attention should go: not the parts that came back for free, but the parts that did not.
This reframes what it means to understand something at all. Understanding is the ability to regenerate the thing from a description short enough that the regeneration took real work. Recognizing a correct answer when you see it is a different and lesser skill, and so is reproducing one from memory. If you can rebuild a proof from its one-line idea, you understand the proof. If you can only recite it, you have memorized a string. The length of the description you can successfully expand from is the honest measure of how much of the structure actually lives inside you rather than on the page.
There is a precondition the cheerful version of "just recreate it" skips, and it is the whole game. Reconstruction only teaches if you control the compression — if you are the one who chose what to throw away, and you discover by failing what could not survive being thrown away. Hand someone both the synopsis and the original and let them glance between the two, and the diff teaches nothing, because the curriculum was never the comparison at the end. The curriculum was the struggle to regenerate with the answer hidden. The skill the post called stealing is more specific and less larcenous than it sounds: it is the discipline of forgetting on purpose and finding out, the hard way, what you could not put back.