The Next Fifteen Years

A forecast built from first principles
Section future / 00-overview / notation.md

Notation - recurring shorthand#


Contents

Terms that carry load across the corpus. Definitions are operational, not dictionary.

TermMeaning hereWhere it matters
Master asymmetryCapability grows fastest where verification / ground truth is cheapData, all of Part III
Ground-truth costCost of knowing whether an output is correct (labels, experiments, P&L, exploits)Domain ordering
Effective computeHardware × dollars × algorithmic efficiency; the real capability engineCompute
Inelastic complementInput intelligence cannot manufacture at will (energy, land, licenses, trust, distribution, physical presence…)Game 3, Assets
Red QueenAdoption mandatory; surplus competed away to customers in competitive marketsGame 3
Apprenticeship gapJunior tasks automated → firms stop hiring juniors → seniors missing in 2040Game 4, Uncertainty 3
J-curveProductivity statistics lag adoption until workflows redesign2028–2032
Leaky bucketFrontier leads diffuse (weights, people, distillation) ~30–50%/yrGame 2
Compute-governance trapExport controls = competitive lever and only future verification tool; using one spends the otherBipolar, Game 2
Salient incidentForcing event that opens an ~18-month rule-writing windowGame 2
Three RSI governorsVerification, physical supply chain, financial (neutral-rate) limits on takeoffUncertainty 1
Two economiesCognition-intensive deflation vs physical/care/energy inflationPrices
Financial governorSuccess raises returns → raises discount rates → dearer next capex roundRates, Capital
Shelf-readinessHaving a drafted architecture before the incident window opensC2
Outcome pricingBilling for results not seats; revealed belief in reliabilityB5, 2026–2028
Two-year moatAny fixed capability level becomes ~free within ~24 months; frontier access is a wasting assetInference economics, Game 3
Revenue bar / capex testDoes AI revenue reach ~$400–700B/yr by 2030, net of financing circularityCompute, Capital, A2
Verification scarcityPost-collapse regime where generation is free and checking is the priced goodGame 5, Compressed, C8
Complement half-lifeScarce-complement rows erode on dated horizons, not foreverUncertainty 7, Assets
Gray-zone TaiwanSoft fail of fab continuity (insurance, licenses, slip) without crisis headlineUncertainty 4, A6
Process vs outcome verificationU5 middle case: process checks learnable, outcome truth still expensiveUncertainty 5, Law
Behind-the-meterCaptive generation / bilateral power routing around public interconnect queuesEnergy, Uncertainty 2

How a term earns a row#

A term is admitted here when it does real work in at least two parts of the corpus and would otherwise be re-derived in each. The definitions are operational on purpose: "inelastic complement" is defined by what happens to its price under demand, not by a category list, because the category list changes (Uncertainty 7) while the operational test does not. When a page uses one of these terms in a way that fails the operational test, that is a bug in the page, not a looseness in the term.

The vocabulary also serves scoring. A prediction phrased in a pinned term cannot quietly migrate its meaning after the fact - "the apprenticeship gap widened" resolves against the definition in this table, not against whatever the phrase has come to mean in 2032. Shared shorthand is a commitment device before it is a convenience.

Three families#

The table sorts into three kinds of object, and the kind tells you how a term can fail. Asymmetries and structural facts (master asymmetry, two economies, leaky bucket, Red Queen) describe gradients; they fail by flattening, and each has an indicator watching for that. Governors and traps (three RSI governors, financial governor, compute-governance trap) describe negative feedback; they fail by being bypassed, which is usually a bigger event than the term itself - a bypassed governor means a branch of Part IV is live. Gaps and windows (apprenticeship gap, J-curve, salient incident, shelf-readiness) describe timing mismatches; they fail by closing early, which is generally good news and is tracked in Part VI as upside skew. A term failing in a way its family does not predict - a gradient that reverses rather than flattens, say - is a sign the framework, not just the term, needs the dependency index treatment.

View markdown source

select · Enter open · Esc close