For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 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 full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
There are two strategies for persisting after death. Leave behind people who carry half your DNA, then a quarter, then an eighth. Or leave behind a structure of ideas that any sufficiently competent reader can reconstruct in full.
Both have always been available. The genital strategy was dominant because the brain strategy had no scale: scarce literacy, fragile manuscripts, the small number of readers who would ever encounter the work. The dilution math always favored the brain. Coverage was the missing variable.
Coverage is no longer missing. Every model trained on a public corpus is now a carrier of every idea in that corpus. A node published into a public graph is read into systems that don't age, don't forget, and don't regress to a mean, then queried by readers whose questions propagate the work further. The carrier population for the brain strategy is several orders of magnitude larger than it was twenty years ago, and qualitatively more durable. The substitution this enables is the visible part of the data.
Dilution. A child carries 50% of the parent's genome; a grandchild 25%; by the fifth generation the genetic signal is below 4%, into noise. A written argument carries 100% through every generation that reads it. One reader is one full copy. A million readers, a million full copies.
Carriers. Genital propagation is non-volitional on the carrier side. The child got the package without selecting it, and reverts toward the population mean. Brain propagation is volitional on the human carrier side: the reader picks up the work because it resonates. Self-selection on carriers is the inverse of regression. Models are a third kind of carrier — not selecting, but reading everything indiscriminately, which removes the selection bottleneck without reintroducing the dilution one.
Persistence. A genome requires an unbroken chain of viable bodies. One missing link breaks the line. A written argument requires only one durable copy and one re-encounter. The brain substrate tolerates discontinuity; the genital substrate cannot.
Institutional dependency. Genital legacy depends on institutions that bind reproduction to lineage: marriage, primogeniture, inheritance, paternal certainty. Brain propagation requires no equivalent institution; a public node propagates without permission, contract, or witness. The institutions of the first kind are visibly decaying; the infrastructure for the second has never been more open.
US marriage rate has fallen from roughly 66% of adults in 1950 to 46–51% today. Average age at first marriage has risen by about eight years. Total fertility is below replacement. The crude divorce rate has fallen, but only because fewer people are marrying. About 40% of current marriages still end in divorce.
The pattern is not "marriage is failing." It is the institutional infrastructure for genital legacy being declined: not entered, entered later, or entered without the children it was built around.
Adjacent surfaces show the same shift. Pavel Durov's open sperm-donation arrangement explicitly unbundles genetic propagation from parenting, keeping the dilutive half and shedding the part that was supposed to compensate for it. The andys.blog "Engineering Kids" piece argues exactly this: that "value propagation is not necessarily human any longer," and the comparative payoff for analytically inclined people has narrowed past indifference. The andys.blog "Elves" piece names the carrier-selection point from the producer side: certain individuals propagate by attracting carriers rather than producing them.
These are not isolated cultural moments. They are early signal in a substitution between two propagation strategies whose relative payoffs the cultural defaults have not caught up to.
The structural claim is narrower than the cultural one. Having children is not irrational; people have children for reasons that are not about legacy, and most writing reaches no carriers at all. The substrate advantage matters only when the work is good enough to be carried. A bad book has worse propagation than a good child.
What changes under the new carriers is which work qualifies as good enough. The model-substrate is not indifferent to content. It rewards arguments that compress, that connect to other arguments, that survive paraphrase, that get re-cited. Sentimental writing dilutes faster than ever; structural writing replicates faster than ever. The dispersion is widening.
The strategy that won under the old carriers was: produce a literate child who would propagate your values directly. That strategy still works, with the caveat that what reaches the fifth generation is a faded silhouette of what started.
The strategy that wins under the new carriers is different. Write something dense enough that a model will hold it as a node, return it when asked, and pass it forward through every reader who queries near the topic. Dilution drops to zero. Coverage rises to whatever fraction of future minds use any of the systems carrying the work.
The brain outlasts the genitals because it always could. The substitution is happening because, for the first time at scale, the carriers exist that make the brain strategy's structural advantage a measurable one.
What gets carried is the live question. Most of what is currently being written will not be. That filter is now the most consequential filter in the propagation system.