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. ↓
The cliff and the decay are the same mistake.
Cliff: Dario at one to three years to AGI, near the end of the exponential. Demis at five from 2026, with one or two breakthroughs needed, ten times the industrial revolution's impact at ten times the speed. Decay: doomscrolling, polarization-is-killing-us, attention has been captured, the kids are not alright, the world feels worse than it did. Most readers hold one of these and reject the other, and feel sober for the choice.
I want to argue that holding either is the same epistemic posture. Both require believing that the long pervasion of computation broke recently: either upward into a near-vertical AGI cliff or downward into a felt civilizational decay. Neither break is in the data. The pervasion is smooth, has been smooth since the lightbulb, and what feels like discontinuity in either direction is the same perceptual error in opposite signs.
The run that is supposed to feel discontinuous: Edison's bulb in 1879, motion picture by 1895, scaled radio by the 1920s, television in living rooms by the 1950s, mainframes through the 1960s, personal computers in the 1980s, internet by the late 1990s, smartphone in 2007, always-on cloud by 2015, generative interfaces by 2022. Continuous, log-linear, with each step shorter than the last and each step folding the previous one into a denser layer. Anyone old enough to remember dial-up did not experience the broadband transition as a phase change. They experienced it as a tightening of an already-running pattern.
The claim is not that any one curve (Moore's law, scaling laws, training compute) is permanently smooth. Specific curves wobble; they slow and re-accelerate; they get replaced by adjacent curves that pick up the trend. The claim is at the level of pervasion: the fraction of human experience touched by computation has expanded continuously, and every replacement curve has continued the trend.
Cutoff predictions ride on the assumption that the next step is qualitatively different from this run. AGI-by-2027, the Demis-style "five years from 2026," even the cooler Dario "powerful AI in 2026 or 2027" framings carry an implicit shape: the pervasion that has been running for a hundred and forty-five years is about to do something it has not done before, namely deliver a categorical event that resets the underlying conditions. The historical record contains no such events. Every previous step looked exactly this way from inside it. Television in 1955 looked like a discontinuity to the people inside that year. By 1985 it looked like one node on the same curve. The view from inside a smooth exponential is indistinguishable from the view from inside a discontinuous one, until enough time has passed.
I am not claiming powerful AI is far. I am not claiming the cliff is impossible. I am making a Bayesian claim about shape. A hundred and forty-five years of evidence licenses a strong prior on continuation; the cliff is the high-variance scenario that requires a layer redefinition above the existing one. The labs running large models are correctly sensing a steep local slope. They are wrong to extrapolate it into a categorical break under the prior the historical record licenses.
Now the other sign. The world feels worse. Polarization is up, attention is fragmented, kids report higher anxiety, institutions trust each other less. The felt sense is real. The question is whether the felt sense matches the underlying world.
Pinker named the perceptual mechanism in The Better Angels of Our Nature (2011). Journalism is an availability machine: the medium samples the worst thing happening anywhere on Earth at any given moment, indexes it for emotional salience, and delivers it as the headline view of reality. Long-run violence has been falling for centuries. The felt sense that the world is on fire is what an availability-driven medium produces when it samples a hundred-million-person attention pool against a billion-event noise floor. The signal that wins is the maximum-bad available signal, every cycle.
The mechanism is now bidirectional. The medium samples the worst physical-world signal globally and delivers it as the headline; it samples the most arresting digital-world signal and delivers it as the texture of daily life. Screens in the car, in the pocket, on the walls. They do not crowd out the physical world. They constitute a second world running in parallel, with its own surplus generation, its own population dynamics, its own real estate where new land is produced at petabyte-per-day rates. People feel the physical world is decaying because they spend an increasing fraction of attention in the digital one, where signal density is higher. The physical is not getting worse in proportion to the felt sense. It is getting less attended to, which feels like decay and is not the same.
The honest version of the story does not pretend the physical record is clean. Three things the cleaner version misses.
First, conflict. Battle-related deaths roughly doubled between 2019 and 2024, plateaued at that elevated level through 2025, and project to remain elevated in 2026. Sudan got worse. Ukraine remained the world's deadliest single conflict. The long-run-since-1945 trendline still bends down, but the recent gradient is up.
Second, productivity. Cowen's 2011 Great Stagnation argues the United States hit a 1973 inflection point: median wages flattened, productivity gains stopped reaching the average household, and the conveyor belt of inventions linking the laboratory to ordinary life slowed. The post-war physical-infrastructure cadence really did slow, and the decay narrative picks up that real signal even when its conclusion overshoots.
Third, compute infrastructure. Building it consumes physical resources at scale, including water, electricity, land, and minerals, in volumes that may not stay small. The two-dimension framing treats this draw as small relative to the surplus generated, and that is the right reading today. It needs explicit accounting if the draw grows.
None of this refutes the smoothness thesis. All of it refutes any version that pretends the physical record since 1973 has been clean.
What these data do not refute is the perceptual point. Air quality in major cities improved through the last two decades. The energy mix in the largest emitters is shifting faster than mainstream forecasts a decade ago expected. Urban centers are revitalizing as autonomous transit erodes parking demand. Nature is returning to areas industrialized fifty years ago. These are slow signals, and they lose every cycle to the availability machine because they are precisely the kind of news that does not generate clicks. The decay narrative wins inside the medium because the medium selects for it. The actual physical record is mixed: improvements outside conflict zones, elevated harm inside them, real slowdown in legacy industries, real acceleration where embodied intelligence has begun routing digital functions into physical processes.
Cutoff predictions and decline narratives both assume that digital and physical compete for the same future. They are different errors of the same shape: treating the new dimension as if its growth must come at the old one's expense.
Digital is a new kind of property. Marginal-cost-zero distribution at the application layer. Composable assets. Attention pools that did not exist in 2002 and now generate measurable surplus: Twitch (2011), Discord (2015), TikTok (2016), the entire creator-economy stack on top of payment rails that did not exist in 2010. New conversation categories: human-to-human at distance, human-to-agent, agent-to-agent, all real conversation, all on this planet, all generating value that did not exist before. The physical world cannot run that pattern; it has friction, geography, atoms.
The two-dimension framing is an approximation that holds at the application layer and weakens at the compute-infrastructure layer, where digital expansion does draw on physical resources. It is useful when the draw is small relative to the surplus; it needs explicit accounting when it is not.
If you treat the two dimensions as one space competing for one attention pool, you get both errors. The cliff is the prediction that digital expansion is about to subsume physical. The decay is the felt sense that digital expansion already has. Both require zero-sum thinking, where one dimension's growth means the other's loss.
Treat them as two and both errors collapse. Digital is where surplus is being generated at the highest rate any historical period has produced. Physical is where surplus is also being generated, more incrementally, on a longer timescale, with real slowdown in legacy domains and real acceleration where embodied intelligence has begun to act. Both compound simultaneously. The polarization-is-killing-us narrative is the panic of a viewer who treats attention as fixed. Attention is not fixed. Two dimensions of it exist now, and they extend each other rather than substitute.
The terraforming wave is the test. Embodied intelligence is already moving from the digital dimension into the physical at scale: Waymo's robotaxi service, Intuitive's surgical platform, John Deere's agricultural autonomy, the new generation of physical-AI startups building manufacturing automation. None of this is fast in the cliff sense. All of it is consistent with the smooth pervasion.
The deeper question is not what the error is, but why it is the default. On a plausible evolutionary frame, humans evolved on a planet where the resources that mattered (food, mates, territory, status) were zero-sum within the relevant timescale. The cognitive default is: another's gain implies my loss. The default was correct for ancestral-physical resources and the right prior for most of human history.
The default is wrong for dimensional expansion. When a new dimension of value-creation opens, the gain is not extracted from any existing dimension; it is generated in space that did not previously exist. Reading dimensional expansion through the ancestral prior produces zero-sum panic, in either of two flavors: "the new dimension will subsume the old" (cliff) or "the new dimension already has subsumed the old" (decay).
The error is not stupidity. It is the pre-installed default running on a configuration that does not match what the world is doing now. Recognizing that the world has shifted from one-space to two-space is the move. The diagnostic that follows is how the move stays installed.
One question.
Does this prediction or this feeling require believing the pervasion broke?
If yes: the prediction requires a categorical event with no historical analog, or the feeling requires that the past was qualitatively better than the present in ways the long-run data does not support. The prediction or feeling is producing a discontinuity hallucination. The discontinuity is in the perceiver, not the world.
If no: the prediction is a steeper local slope on the same curve, or the feeling is an honest partial sample of a multi-dimensional reality where one dimension is genuinely accelerating. The prediction or feeling is calibrated.
The diagnostic is a Bayesian filter, not a categorical refutation. The cliff is not impossible; layer redefinition above an existing one is a real category and has happened before. The decay is not entirely fictional; recent conflict and Cowen-flagged productivity data carry real signal. What the diagnostic refutes is the high-confidence version of either: the AGI-by-2027 confidence, the civilization-is-collapsing confidence. Both read discontinuity into a smooth-but-fast pervasion. Most cutoff predictions fail the diagnostic. Most decline narratives fail it for the same reason in the opposite direction. The world is on the same slope it has been on since 1879. Nothing in the present requires believing that slope broke.