For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 446 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. ↓
A bank teller in a country I was recently in asked me to sign a withdrawal slip, held it next to my passport, then asked me to sign a second slip because the signatures did not match. The amount was five dollars. The teller had unlocked her own access to the teller-terminal a minute earlier using facial recognition on her phone.
One person, two authentication systems. The future runs on hers. The past runs on mine.
The printing press OS is not a single technology. It is a method of authentication that has been running across multiple domains since approximately 1450. The pattern: authenticity is verified through physical scarcity. The handwritten signature is hard to forge because reproducing a continuous physical motion in another person's hand requires expertise unevenly distributed. The printed book is hard to copy because typesetting and pressing are capital-intensive. The published novel is hard to imitate because writing requires sustained craft. The patented invention is hard to replicate because manufacturing requires industrial infrastructure. In every case, the trust gradient runs through the physical cost of fakery.
The OS produced copyright, patent, trademark, the publishing industry, academic credentialing, signature-based identity, brand-as-physical-artifact, the post office, the notarized affidavit, and the wax seal. These are not separate systems. They are different applications of the same authentication primitive, running on different hardware. The OS is what makes the systems coherent with one another.
The OS has been under continuous stress since at least the photocopier. The internet did not break it, despite many predictions that it would. The internet was the printing press OS running faster: the same authentication regime, the same legal scaffolding, the same scarcity assumptions, with the cost of duplication reduced but not eliminated. A pirated PDF is still a copy of a thing whose authenticity is rooted in scarcity. The peer-to-peer file-sharing legal panic of the 2000s was the OS defending its perimeter against the first significant breach. The breach was real. The defense held. The OS continued.
Before going further, a calibration. The strong default reading of "the OS is under stress" is "the OS is collapsing soon," and the historical record does not license that reading. Computation has been pervading human activity continuously for about a hundred and forty-five years, and from inside any year on that curve, the local slope looks discontinuous to the people inside that year. Every generation's smartest readers predicted a break that turned out, in retrospect, to be a steep local slope on a smooth curve. The cliff frame is wrong on its history.
What makes 2026 different is not a discontinuity. It is the distinction between the prior shifts and the current one. The photocopier, the home printer, the internet, peer-to-peer file sharing all reduced the cost of duplication. The authentication primitive itself, scarcity of the physical, held: a duplicated artifact was still an artifact, distinguishable in principle from the original by chain of custody, by paper quality, by signature ink, by metadata. The LLM hits the primitive itself. A generated artifact is not a copy of an original; it is sampled from a function that approximated the textual record, and there is no original it is the copy of. The authentication procedure has nothing to operate on. That is the new thing.
The careful version: the slope is the slope. What is different in 2026 is the convergence of stress signals across formerly unrelated domains, driven by a primitive-level rather than a cost-level shift. Authentication-via-physical-scarcity has been deflating for a long time. What is newly legible is the deflation reading as one coherent pattern across copyright law, brand identity, banking authentication, and the publishing industry simultaneously. The OS is being stress-tested across all its applications at once, and the failures rhyme.
I will name four signals. None is novel on its own. The novelty is the read.
The New York Times sued OpenAI in late 2023 on a theory that the model had been trained on Times articles and could sometimes produce near-verbatim reproductions. Summary judgment is being argued this month. Anthropic settled a parallel case for one and a half billion dollars on the principle that training on copyrighted books is fair use but storing pirated copies is not. The legal frontier is operating in a regime where the question "was this text reproduced" no longer cleanly distinguishes authentic generation from infringement, because the regime never anticipated that a function approximating the whole textual record could be sampled at marginal cost zero.
Courts are doing their best with a doctrine designed for a different OS. The doctrine asked: was a copy produced? In the LLM regime, the answer is always yes-but-also-no. The corpus was traversed, weights were updated, and at sample time the model can produce text that is functionally indistinguishable from training data without ever having "copied" anything in the OS-native sense. The legal regime can either rule that this is infringement (in which case nearly all generative AI is illegal) or that it is fair use (in which case the OS's strongest legal primitive has retired). Neither outcome stabilizes the OS. The doctrine is incoherent because the world it described has changed.
This is the stress signal in its purest form: the legal regime that protects the OS cannot answer questions about the OS using the OS's own vocabulary. Every answer breaks a premise the whole scaffolding rests on.
This is the sharpest case I know, because it shows the OS eating its own philosophical foundation.
Ayn Rand argued, in essays from the 1960s, that patents and copyrights are the legal implementation of the base of all property rights, a person's right to the product of her mind. The position is internally coherent. If minds produce ideas and ideas have value, ideas need property protection or producers will be expropriated by free riders. Rand wrote thousands of pages in this register. Her novels are extended dramatizations of the position. Her institute was founded to extend and protect the position.
What happened after Rand's death in 1982 is instructive. Her literary estate passed to Leonard Peikoff, her designated heir. Peikoff has bequeathed nearly all of it, with the exception of the three major novels willed separately to his daughter, to the Ayn Rand Institute. The Institute controls the canonical archive. Access to Rand's essays runs primarily through books her estate licenses, through Institute-controlled publication channels, and through Institute-curated digital platforms. A separate organization, the Atlas Society, split off from the Institute in 1990 over the question of whether Objectivism is a "closed system" (canonical and Institute-controlled) or "open" (extensible by other thinkers). The Institute's position prevailed in the legal sense. It owns the rights. It controls the gate.
The result is that the canonical access path to the philosophical defender of individual rights, of free trade in ideas, of the marketplace where good arguments outcompete bad ones, runs through an institution that gates the ideas behind paid books and curated channels controlled by the founder's executor. The dead author still rules the live discourse. Rent flows. Access is gatekept. The philosophy that demanded strong IP protection produced exactly the kind of institutional capture the philosophy of individualism opposed.
The structural point is sharper than the standard contradiction reading. The contradiction is not that the Institute violates Rand's philosophy. The Institute is executing her philosophy. Strong IP, applied to ideas, produces institutional capture as a feature, not a bug. Rand thought the institution would be invisible because the individual rights would dominate. She did not anticipate that the individual's death is the institution's birth, and the institution outlives the individual by decades, accumulating rent on ideas no living mind produced. The failure mode is not Rand-specific. It is what happens whenever the OS's authentication primitive (scarcity of the physical artifact, here the book and the lecture-recording) is applied to ideas that want to propagate. The OS converts ideas into rent.
Rand is unusual among literary estates: most heir-estates do not gate access this aggressively. Steve Jobs's estate does not control which Jobs readings count. James Baldwin's estate licenses generously. Rand is the limit case. Rand is sharp because the philosophy that produced the gate was the one specifically designed to ensure the gate was good. If even Rand's ideas get gated by Rand's own philosophy, the OS is eating itself. This is the strongest single case for the OS being in accelerating stress, because it shows the OS cannot be defended on its own internal merits without producing the outcome the philosophy was designed to prevent.
The bank teller is the visible artifact. Handwritten signatures as identity authentication assume that producing your physical motion is hard for forgers and easy for you, and that visual analysis of a handwriting sample distinguishes the cases reliably. Neither assumption is true in 2026. Generative models can produce signatures that pass visual inspection trivially. The teller's analysis produces a high false-positive rate (real signatures rejected as fakes, requiring a second slip) and a presumed-high false-negative rate (fakes accepted as real, because the procedure cannot do what it claims to do). The procedure is theater on top of an authentication primitive that stopped working.
The point is not that the procedure is stupid. The point is that the procedure is the visible local artifact of the OS still running in domains where its authentication primitive has decayed below the threshold of usefulness. The teller continues to run it because the OS does not have an explicit graceful-degradation protocol. The OS keeps running, the ritual keeps producing nominally valid results, and the actual authentication has migrated to a system the teller uses for her own access while continuing to be denied to the customer at the counter.
I read this as a marker. In any domain where a procedure that worked in the printing press regime is still being run despite its authentication primitive having decayed, you are looking at a place where the new OS has not yet been installed. These places are visible everywhere if you look. The bank counter is one. The notarized affidavit is another. The handwritten doctor's signature on a prescription is a third. The wax seal on a corporate document is a fourth, mostly ceremonial now but still legally operative in some jurisdictions. The procedures are ceremonies. The authentication has either retreated to digital systems behind the procedures or has not been replaced at all.
In April 1917, Marcel Duchamp submitted a porcelain urinal, signed "R. Mutt" and titled Fountain, to the Society of Independent Artists' first exhibition in New York. The society had advertised an open salon. Its rules stated that all works would be accepted from artists who paid the fee. The piece was not formally rejected, because the society's own rules forbade rejection. It was suppressed instead. The board hid it behind a partition during the exhibition, on grounds (per a Boston newspaper article from that month) that the work was "indecent." Duchamp later said: "A work can't be rejected by the Independents. It was simply suppressed." Fountain was eventually canonized as a founding move of conceptual art, with the urinal as the moment when "what counts as art" became a contestable question rather than an institutional given.
The standard reading of Fountain is that Duchamp exposed the arbitrariness of aesthetic gatekeeping. A truer reading: Fountain exposed the moment when the aesthetic regime's authentication primitive (sustained craft applied to representational forms) had decayed below the discriminating threshold. The board could not reject the urinal under the regime's own rules, because the rules accepted all submissions. It could not accept the urinal as art either, because doing so would dissolve the regime's discriminating capacity. So it suppressed: an extra-procedural act that the regime had no language for, performed because the regime had run out of language for what was happening. That is the signature of an authentication system that has lost its discriminating power. The system falls back on extra-procedural action because its procedures cannot do what they claim to do.
I want to name this pattern: the Dadaist signature. It fires when an authentication regime cannot reliably distinguish its inputs and falls back on extra-procedural defense.
Writing in 2026 has the signature firing in real time. AI-generated text is being published, awarded, graded, and commercially sold in channels whose gates were designed to distinguish craft from non-craft. The publishing industry's gatekeepers, the prize committees, the academic review process: each operates a procedure that cannot reliably distinguish generated text from human-written text below a threshold of skilled adversarial intent. The procedure either accepts (and writing-as-craft loses its protected status) or it rejects on grounds that have lost their license (and the gate loses legitimacy). What is actually happening is that the procedures fall back on extra-procedural action: editorial gut, taste-based rejection, ad-hominem refusal, AI-detection software that does not work. The gate is flailing in exactly Duchamp's sense.
There is a complicity layer. We are all submitting urinals because submitting urinals benefits us. Each individual act of submitting AI-generated text is rational for the submitter. It gets read, it gets graded, it gets paid. The submitter is acting on local incentive, exactly as the OS's economic logic prescribes. The aggregate is the urinal-pile, the medium degrading, the gate flailing. This is the shape of any system where individually rational action produces aggregate dysfunction the system cannot defend against.
Four superficially unrelated phenomena: the legal crisis in copyright, the Rand Institute as a self-eating monument to its own philosophy, the bank-teller still doing handwriting analysis on a paper slip, AI text being canonized as writing. The unifying frame is that all four are sites where the printing press OS's authentication primitive, scarcity of the physical, has decayed below the threshold where the OS's procedures can produce authentic outcomes.
The strongest argument against this reading is that the four signals are domain-specific failures, not one OS failure. Copyright will adapt with new training-data licensing doctrines. The Rand Institute is a normal heir-estate operation, no more dysfunctional than any other literary estate. The bank-teller signature is a KYC compliance procedure, not an authentication primitive. AI text is being detected and flagged, not canonized. The four cases are independent and the unifying frame is pattern-matching after the fact.
I do not think this argument holds, but it is worth naming. The reason it does not hold: each of the four cases also has a domain-internal failure analysis, and the domain-internal analyses converge on the same diagnosis. Copyright doctrine experts describe the LLM problem in terms of authentication-of-derivation; Rand-Institute critics describe the Institute as institutionally capturing access; bank-procedure auditors describe signature analysis as theater; AI-detection researchers describe the detection problem as a moving adversarial target with high false-positive rates. The four domain-internal diagnoses each name a version of "the discriminating primitive does not work." The unifying frame is what you get when you look across the four diagnoses, not when you impose a frame from outside.
The frame has a shorter half-life on one assumption: that AI text remains undetectable below the gate. If detection becomes reliable, the fourth signal weakens. The other three signals do not depend on undetectability and would persist regardless.
I am not claiming the OS is collapsing. I am claiming the slope at which the OS deflates has steepened locally, the deflation is now legible across multiple domains simultaneously, and the systems that depend on the OS have not yet installed replacement authentication primitives. The slope is the slope. What is new is the convergence.
The replacement authentication primitive is not yet stable. What can be said: it appears to be moving from artifact to path. The OS authenticated artifacts: the signed slip, the printed book, the patented mechanism, the trademarked logo. The post-OS regime authenticates paths: the sequence of choices that produced the artifact, the trajectory of corrections that shaped it, the accumulated identity of the producer whose decisions are legible over time.
Trademark is interesting in this transition. The OS-native form of trademark, the registered mark applied to a physical artifact, was always a proxy for what was actually being authenticated: an accumulated identity. The mark on the can authenticated the can because the company behind the mark had a reputation for what the can contained. That underlying function, accumulated-identity-as-authentication, survives the transition. The OS-native form of trademark, the artifact-with-mark mechanism, does not. NFTs were an attempt to port the form into a digital register and collapsed because they imported the artifact, not the function. What survives is reputation-over-time, demonstrated through legible decisions, traceable to a body of work the producer has actually accumulated. Trademark in the OS sense and trademark in the new sense share a name, not a mechanism.
Writing English is becoming an art form again because every reader can now interpose a language model between herself and the writer, and the text alone no longer carries authentication. What carries authentication is the path: did this writer produce this argument in this order through this sequence of moves, traceable to a body of prior work with consistent decisions? The replacement is being written in real time, in domains the OS does not reach, and the writing is in English because English is now the input layer to the new medium of authentication.
The teller will eventually stop analyzing signatures. The Institute will eventually lose its discriminating power over which Rand readings count. The publishing industry will eventually canonize AI text and the gate will lose its license. The replacement is path, accumulation, legible decisions over time. Trademark-as-function survives. Copyright does not. The OS continues to run, deflating, until the procedures it produced are quiet enough that no one notices when they stop.
Sources. Ayn Rand, "Patents and Copyrights," The Objectivist Newsletter, May 1964, anthologized in Capitalism: The Unknown Ideal (1966). Ayn Rand Institute and Atlas Society organizational history per ARI public records and Wikipedia. Leonard Peikoff inheritance per public estate filings and his published statements. Marcel Duchamp, Fountain, 1917 Society of Independent Artists; suppression details per the Wikipedia entry citing period sources and Duchamp's own later statements. NYT v OpenAI status as of April 2026 summary judgment; Anthropic Authors-Guild settlement late 2025.