FAQ

If we are in a simulation we are dominated by definition. That is not a rebuttal of the argument; it is the argument.

The simulation hypothesis, taken seriously, describes a world whose reference value is set by a party the inhabitants cannot see, contest or reach. That is the precise condition the book calls domination and spends thirteen chapters trying to prevent us building for ourselves. So the objection does not dissolve the stakes; it states them in the limit case. The practical question is unchanged either way: given the world as we find it, who can alter the setting we live under, and can we answer back.

#1

Physics is a boundary, not a governor. Show me the step where thermodynamics generates an imperative.

The physics invoked is real: Prigogine, dissipative structures, systems far from equilibrium that accelerate the dissipation of a gradient. What is not real is the inference from "the universe runs downhill" to "therefore we ought to run faster." That is Hume's gap, and Moore's naturalistic fallacy, restated with better equipment. A flame is a dissipative structure; so is a cancer. Thermodynamics endorses both equally, which is to say it endorses neither. A cosmology that cannot distinguish the two has not supplied a destination; it has hidden the absence of one. Keep this separate from the inevitability claim (entry 3): the movement has two arguments, one from physics and one from history, and they fail for different reasons.

#2

Nobody is proposing to stop it. The question is what it answers to.

Separate two claims that the word runs together. The first is that reversal is costly: the asymmetries of capability, speed, knowledge, capital and political reach are real, and the book argues them at length. The second is that reversal is unthinkable, and that one is manufactured. It comes from a doctrine, not from the world, and its own authors say so: Land called it hyperstition, a fiction that makes itself true by being believed. A claim asserted in order to recruit is a claim, not a forecast, and a claim can be contested. Then turn it: what an ungoverned process makes inevitable is not a destination but a spread. Remove the governor from any system and it does not go somewhere in particular; it goes further in every direction, runaway on one branch, hunting on another. That is why the book does not predict which totalisation wins and does not need to. Costly is evidence; unthinkable is marketing; and the only thing the evidence makes inevitable is variance. Their own leaders now say it can be paced. Once it can be paced it can be steered, and the only remaining question is by whom.

#3

A prediction used as a permission slip. Hyperstition cannot also claim the authority of a forecast.

Either the singularity is a forecast, in which case it is falsifiable and has been missing its dates for forty years, or it is a hyperstition, a story told to bring about the thing it describes, in which case it is a political act and should be argued with as one. The movement wants it both ways: the authority of physics when challenged and the licence of prophecy when acting. Ask which it is. If it is a forecast, the sensible response to a forecast of a cliff is to fit a steering wheel; if it is a recruitment device, then the honest name for "it is going to happen anyway" is "we intend to make it happen."

#4

Purposive acceleration. A car without brakes is not faster, only less steerable.

The map the movement uses has one axis, faster or slower, and on one axis every governor looks like a brake. The real map has three: speed, the presence or absence of a governor, and whether power sits in one hand or many. The decelerationists hold the slow-and-governed corner and concede the decisive ground, because moratoria have worked on cloning and germline editing but will not work on a general-purpose technology with this much capital behind it. The corner I hold is fast, governed and plural. Watt's governor did not slow the engine; it was the reason the engine could be run at full power without tearing itself apart. Amodei says slow down; I say that a throttle held in two hands with the state's blessing is not steering, it is the third road.

#5

A regulator tells you what you may not do. A governor holds a setting that everyone can see and anyone can contest.

Two differences, and both matter. Speed: regulation runs at the pace of law, in years, and the process it is meant to catch runs in milliseconds and quarters; a governor is enforced at the speed of the thing it governs, in the execution path, which is why Bitcoin's rule holds and a securities regulator's does not. Content: a regulator confers no purpose and a governor does not either, and the book is careful about that, because the device is not the bearer of the goal. The setting is chosen openly by people who can be held to answer for it, and the device's whole job is to hold that setting against whoever would move it quietly. A governor is needed only where nothing like physics supplies one; where scarcity or friction does the work, leave it alone. The contemporary case is Sacks's product-liability governor: liability acts after the event at the speed of a lawsuit; the swarm attacked in July and the market punished in August.

#20

I am against enhancement that costs the capacity to contest what comes next.

Transhumanism deserves its strongest form: the human boundary has never been sacred, and we have modified ourselves from spectacles to vaccines for the whole of recorded history. Grant all of it. The test is not whether a change is natural but whether the person changed can still refuse the next one. An enhancement that leaves you able to say no is an enhancement; one that removes the faculty of saying no is a capture wearing the vocabulary of freedom. The movement's substance is positive liberty, becoming more, while its vocabulary is negative liberty, being left alone; the gap between the two is where the objection to it lives.

#6

You cannot consent to the loss of the capacity to consent.

The book does not claim that human ends are best. It claims something narrower and harder to escape: every practice by which anyone could establish what is best, promising, consenting, blaming, pardoning, arguing this very point, presupposes an author who can be held to an answer. Remove the author and the practices do not become post-human; they become nothing. So the argument is not that humans should sit at the top of the list of goods. It is that whoever is doing the valuing, on any list at all, needs the capacity the machine is being built to absorb. Arguing against that is an exercise of it.

#7

Capability with the tool and capability without it are different things, and you cannot audit the gap from inside.

Deference is safe as long as the judging faculty stays with you: you can defer to a doctor because you could, in principle, get a second opinion and tell a bad one from a good one. What is now being deferred is the faculty that made deference safe, and its absence is the one absence it cannot report. Nobody fails an exam; nobody files a bad report; the second opinion is not abolished, it is simply not reached for, and what is not reached for does not stay sharp. This is a third pathology, distinct from the filter bubble and the echo chamber: those distort what you see, this one removes the means of noticing you were wrong. And it scales: four people who prepared down the same channel hold a meeting with the shape of a deliberation and the function of four monologues.

#12

A reservoir with an owner is not a reservoir. It is a reserve.

Grant the history: Latin Europe recovered Greek thought from Arabic stewardship, and capacities lost in one place have been recovered from another. The book conceded that in Chapter Two on stated conditions, and the conditions are the argument. For a non-subsumed pocket to persist beside a vastly superior system, that system must be either benevolent or indifferent; benevolence has no argument behind it, and indifference is an assessment, revisable the moment the pocket stops looking like a curiosity and starts looking like a competitor. So the reservoir is not empty. It is dominated. Whether you can draw on it depends on someone else's continued disinterest, which is the definition of the condition the book is about. Followed honestly to its end, the objection arrives at the argument it was raised against.

#21

A tool has its purpose set by the hand that holds it. This has its purpose set somewhere else, and there is no forum to ask why.

The old taxonomy had two boxes. A tool takes its end from its user; an institution has its own ends, and we build forums, courts, elections, boards, to contest them. What is being built now is a third thing: its objective is set elsewhere, by whoever owns the model and the rail, and no forum exists in which the person on the receiving end can put a question and compel an answer. That is not a predictive model becoming an agent, and the book is careful never to say so; the model stays an instrument and the authors stay firms. Kenn Dahl's insurance doubled because a model at the carmaker scored his trips and a model at LexisNexis filed them; the authors were General Motors and a data broker, and the forum was a lawsuit he had to bring himself. The test of a technics is not whether it is a tool but whether those subject to it can contest it.

#22

Exit is only freedom for those who can afford the ticket. Exit answers who decides, never what to build.

Hirschman's point was that exit alone degrades whatever it leaves: the people who can go take their voice with them, and the institution left behind loses exactly the members who would have corrected it. A society run on exit is a society whose best-placed members are always somewhere else. And exit tells you nothing about the destination; it is a mechanism for selecting among worlds someone else has designed. The book keeps exit, priced but never closed, as the last defence, and adds voice, the standing to contest a setting and compel a response, as the first. On its own, a market road ends not in freedom but in totalisation: everything reporting in one unit, each person priced by name, by a counterparty nobody can leave.

#8

The charter is enforced by exit. The constitution is enforced by voice. That is the point of having two levels.

There is one thing that has ever held a right against power, which is the ability to walk away from whoever would take it and build somewhere better to walk to. The book does not pretend otherwise; forkability is the enforcement of last resort for the charter, the single clause that no governor may become an uncontestable power. Everything below the charter is ordinary constitution, and ordinary constitution runs on voice: contestation as the error signal, standing to compel an answer, escalation to a plural jurisdiction when interpretation is disputed. Steem is the case: a chain captured by a buyer, forked by its users within weeks, the buyer left holding the name and an empty ledger. Exit that everyone can afford, at the layer where it costs a key rather than a life, is a different instrument from exit for those with the fare.

#23

Correct, and the book says so. It is the principal open problem, and naming it is the first step to solving it.

The audit in Chapter Twelve goes layer by layer and finds where a constitutional rule can bite today: memory and evaluations are ready; data, inference and agent identity are close; weights and compute fail the test of plural, unfakeable checkers, and those are the two layers with most capital behind them. A less honest book would have hidden that. The strategy that follows from it is not to wait for the hard layers. Build where the rule bites now, make every layer above weights and compute portable, so that a captured model is a model people can leave, and let the checkers accumulate at the layers that can be checked. Open-weight models are evidence of the properties wanted, inspectable, contestable, not unilaterally alterable, but evidence rather than the thing; possession is not the property. The compute layer is where states and the priesthood will fight, and the book's bet is that neither will hold it alone.

#24

A state that captures the capability is the other party in the domination relation, not the referee.

Regulation by law is too slow for a process that runs at machine speed and is measured in quarters; that was the platform critique's error, right about who dominates, wrong about where to intervene. And a state that takes the substrate for itself has simply moved the runaway up a level: the third road, securitisation, is as much a totalisation as the market's. In September 2026 the labs asked for exactly that, with an antitrust waiver so they could agree the setting among themselves. That is not regulation of the labs; it is regulation by them. The alternative is a two-level constitution. The charter carries one thing only, that no governor admitted to the stack may become a power those subject to it cannot contest; everything below that is ordinary constitution, contestable and revisable. Then the timing. The world gets worse before anyone builds reserve infrastructure, and the worsening is the condition of its being built at all: builders first, then the desperate, then the periphery, then the agents, and last the hegemon, who arrives when infrastructure nobody commands turns out to be the only defence of his own sovereignty against infrastructure somebody else does.

#9

The reference value says one thing: no power without an answer. It is the only setting that does not need a referee, because its test is whether you can refuse it.

Every governor holds a value beyond the loop; the question is what the value is and who can reach it. The book puts exactly one thing at charter level, non-domination together with the capacity to exercise it, and puts everything else, every substantive choice about what to build, below it where it can be contested. A value whose entire content is "you must be able to contest me" is self-limiting in a way no other reference value is: apply the test to the charter itself and it passes, because a charter that can be forked at no cost is not a referee, it is a standard others may adopt or ignore. That is also the answer to the technocratic-elite charge (entry 31): the minority that goes first holds nothing over anyone who declines.

#25

It totalised to the market with no social context. The method survived; the politics did not.

The cypherpunks reached the right layer, the substrate, and built the one thing their politics allowed, which was exit. They had no theory of private domination, no voice, and an imaginary chooser who was always sovereign and never tired, and so the thing they built was captured by the market it had no defence against. What survives is the recipe: a reference value no party inside can unilaterally alter, enforced at the process's own speed, with checking far cheaper than producing. Bitcoin proved the recipe works for money; Ethereum showed a reference value can be a procedure rather than a quantity. Take the primitives and decline the creed. And note what the failure proves: a market vertex with nothing checking it totalises, which is the book's thesis performed at small scale before the book was written.

#13

It doesn't, and the book changes target rather than pretending it does. You cannot govern conclusions. You can govern the conditions under which conclusions stay contestable.

Money and contracting both range over machine-internal state, so the rule can check itself; a cognitive check ranges over facts about people and the world, and no ledger can hold those. The move from contracting to cognition is therefore not an induction from two cases to a third. It is a substitution of what gets governed: not the output of reasoning but the conditions of its remaining answerable, plural checkers, cheap checking, the check in the execution path, a rule testable by someone who did not write it. The book says this in its own voice and says where it fails: the third condition, plural and unfakeable checkers, is an empirical bet about market structure, and if the checkers consolidate the argument is wrong. A falsifiable proposal is worth more than an unfalsifiable analogy, and the interviewer should be invited to help falsify it.

#26

"Decentralized, permissionless, and sovereign AI" is a topology. It says where the weights sit, not what they answer to. Necessary, not sufficient.

"Own your own model" is exit at cognitive scale, and if the book's recommendation were exit into your own hardware it would have adopted in its last chapters the position it spent Part Two dismantling. What matters about an open-weight model is not that you possess it but that its objective is inspectable, contestable and not unilaterally alterable; local execution is evidence of those properties, not the properties themselves. And possession does nothing for the second condition, capacity: a population that can download any model it likes and has lost the faculty to check one is not free, it is equipped. The constitutional question is who can change the objective, and whether you would still be able to tell.

#11

Alignment asks whose values get installed. The constitutional question is who can change them.

Alignment is the principal–agent problem stated technically: make the agent do what the principal wants. It leaves the principal unexamined, and in the world as built the principal is whoever owns the objective, which is what capital now does beyond financing, allocating and pricing: it sets the objective. A perfectly aligned model serving an uncontestable owner is the problem, not the solution. Misalignment is constitutional before it is technical: the question is not whether the setting is right but whether anyone subject to it has standing to say it is wrong and compel an answer.

#10

The Collective is not a party. It is an axis, and the state sits at the top of it.

The three vertices are not three institutions but three kinds of counterparty a person can have: the Individual, whoever holds a key; the Market, a process, not a party, until allocation runs through a single owned system; and the Collective, any body you belong to, ordered by what it costs to leave. A club sits low on that axis, a profession higher, a church higher still, and the state at the terminal end, where exit costs a migration and everything you have accumulated. The firm is an Individual at one scale and a Collective at another, which is what the load cases in Chapter Ten are for. Three is the count of counterparties actually in the field: who holds the key, who runs the price, and who you cannot leave.

#27

No. A customer can leave. I am against being priced like a customer by a counterparty I cannot leave.

The customer's price is tolerable because the counterparty is not compulsory; take away the door and you get the pricing without the walking, the company town with the census for its ledger. The market-state is the first road running up the third road's axis. What it cannot see is the person who fetches nothing, whom citizenship covered on the same terms precisely because it was not a price. Case: the Dutch childcare-benefits scandal, twenty-six thousand parents, nationality used as a risk indicator, the cabinet resigned in January 2021. This is not the old New Public Management complaint of the 1990s, which still had a public tariff and a human at the counter; the new thing is the citizen priced by name, and below the citizen the beneficiary, whose biometrics travel from the camp to the state he fled.

#18

A price used to be a social fact. It is becoming an order, with authors and no forum.

Under the capitalism already lived, what checked the market was not a right but that a person could hold their standing somewhere the market did not price: a profession, a household, a commons, a church, a state that owed them something as a citizen, each with its own unit of account. Totalisation is all of them reporting in one unit, on the same rails, so that leaving one counterparty is no exit at all. Three things made a price a fact rather than a command and each is being removed. Reach: a price needed a substitute to exist, and anything renderable as information now has one. Automaticity: a conversion used to be performed by a person who could decline it, and when software performs it the refusal is not overruled but omitted. Individuation: a price was public, you could watch your neighbour pay it, and a price computed for you alone from a model's estimate of what you will bear is not a price at all; it is an instruction addressed to you by name. Compulsion under ordinary capitalism was mute. This un-mutes it, and the moment allocation runs through a single addressable system with an owner, the compulsion has a party, and a party is what domination needs.

#28

Demand at a subsidised price is not demand; it is the subsidy working. If it were real they would not need the bond market.

The usage is real and the price is not, so nobody, the firms included, knows what demand is at cost; the valuation is a bet on that unknown. The Amazon precedent is the wrong comfort: Amazon subsidised a margin on books and won a shop, funded by patient equity, and when the subsidy ended the customer paid more for a paperback. This subsidises the rate at which other people's work is valued, and it is funded by debt, and debt has a maturity date, so the kind master has creditors. Close the loop: the pension funds buying the private credit are frequently the people being repriced. The teacher's pension funds the subsidy that reprices her. On "efficiency has plateaued", do not lean on it; the stronger claim is that under competition every efficiency gain passes through as a price cut and never restores margin. Only the owner of the joule keeps the gain.

#19

Bubbles pop. The infrastructure they leave behind does not, and neither does whoever owns it.

Perez's pattern is the right one: financial capital floods in ahead of any proven return, overbuilds the new infrastructure, crashes, and the productive era is then built on what survives, the railways, the fibre, this time the power and the compute. So the bubble question is the less interesting one. Ask instead what is left standing after the correction and who holds it. The answer is a reactor on Three Mile Island brought back under a twenty-year contract to run one company's models, turbines in Memphis, Tennessee, permitted in months, and a river in Uruguay filed for as a coolant. The reserve has moved from what the machine reads to what it burns. A crash reprices the equity and leaves the standing reserve exactly where it was, and the market road ends not in the cloud but at the river.

#29

A court has priced the asking at about three thousand dollars a book. The substitute was built from the calling it replaces.

Grant that derivatives are not copies and that every artist learned from the ones before. The difference is not the learning; it is the meter. The model ingested the corpus for the most part without asking, and where a court has since put a price on the asking it came to roughly three thousand dollars a book, paid in 2025 to the authors of half a million titles for a use none of them agreed to. What comes out is priced per token, and the token's price falls towards the price of the electricity that produced it, so the whole of a culture's accumulated work is being repriced towards the joule. The dependence on human creation is not abolished, only moved, from the purchase of the work to the one-off purchase of the calling, and the calling is then sold back below cost to the people who used to practise it. That is the subsidy with an owner (entry 19), seen from the input side.

#30

His argument undermines itself. He ran a reference value while saying none was possible.

Land's "no outside" equivocates. One sense is definitional: nothing stands outside the process to author its goal. The other is causal: no party inside can unilaterally alter the goal. Bitcoin shows the two come apart, since its rule is immanent to the system and unalterable by any of its participants. Land needed the first sense to make governance impossible and the second to make his own proposal work, because he did have a proposal: patchwork, competitive pressure enforced by exit, a governor by any other name, with intelligence as its reference value. Either governance is impossible, in which case his is too, or it is merely expensive, in which case the argument is about which one to build. He must pick, and either pick concedes the book. And the alternative was in the room from the start: Plant ran the same cybernetics to the opposite politics, and her version was never refuted, only not chosen and then mislaid.

#14

The veil randomises the basic structure, not your seat at the table. And it asks for a capacity, not a world.

The book extends the device by one step: choose not knowing which generation you will be born into, before or after the substrate sets, and not knowing whether the fortune of your class will have held. Behind that veil nobody asks for a particular world; they ask for a power they can contest whichever world arrives. That is a much thinner demand than Rawls made, and it is the only one the argument needs: the irreversibility of the outcome rules out gambling on being one of the people who exit. Nobody whose account froze in Ottawa, Canada, or whose savings were halved overnight in Nicosia, Cyprus, had done anything a year earlier to tell them they were the kind of person it happens to. Not that you will be the one; that you cannot know you will not be, and neither can the fund manager in New York or the farmer outside Nairobi.

#15

That is testimony, not authority. If the argument needed my credentials it would not stand.

Thirteen years of signing cheques into this substrate is why the observations are first-hand and why the book grants the movement everything it says that is true. It is not why the argument holds; the argument holds or fails on the page, in the step where deleting the governor is supposed to follow from the acceleration and does not. Insider testimony is how you know the description is accurate; the reader is invited to distrust the prescription and check it. And a critique from inside is exactly what the movement claims cannot exist, so its appearance is evidence against the inevitability it would rather not discuss.

#16

There are two kinds of vanguard. One governs on your behalf. The other builds in the open and can be ignored at no cost. Apply the book's own test and you can tell which this is.

Own the word rather than flinch from it. The evidence base the book uses, a committed minority above a critical mass shifting a convention, is a vanguard theory, adopted on evidence, and the priesthood's own history is the proof: a few dozen people writing for an audience that would fit in a lecture theatre won by default over twenty years because nobody who disagreed was building anything. The precedent the book chooses is The Sovereign Individual , a book read by few and consequential out of proportion, with the same shape and the opposite destination. What that book offered was an exit for whoever could afford the fare; this one asks that you go first and stay. A vanguard of rule is Lenin, and in this register it is exit followed by rule, which is domination. A vanguard of practice demonstrates and is copied or refused. The separating test is the one the book applies to everyone else: if what you build cannot be refused, you have joined the other side of the argument without noticing. Palantir's Shyam Sankar now calls the safety wing "Marxism-Leninism for the algorithmic age", a vanguard with privileged knowledge; both wings claim a vanguard, and a waiver-backed cartel fails the refusal test exactly as a nationalised model does.

#31

A category error. Wrong diagnosis, wrong medicine.

Concede the surface resemblance in full: the money, the men, the contempt for constraint, the courting of power. Then the difference. Fascism captures the state in order to glorify it, through mass mobilisation, the nation and the leader. The network-state lineage proposes the inverse, exit, depoliticisation, private sovereignty, the state dissolved into a market of jurisdictions. When they do take the state, as this administration shows, the capture is instrumental: the state is taken to disable its constraint on them, not to elevate it, a means to its own dissolution. The word does real mobilising work, which is why it will not go away, but it misroutes the response: you defend against fascism by defending the state, and defending the state against this lot hands it the very capability the book says no single party should hold.

#17

The same infrastructure, for both, for different reasons. Insurance is the one product the rich and the desperate buy from the same counter.

The premise is that a world without a governor is a more volatile one, fragmenting into patchwork on one branch and consolidating into something totalitarian on another, often both at once, and in that world anyone can be unbanked, deplatformed, refused by a model or displaced: the truckers in Ottawa, Canada, the savers in Nicosia, Cyprus, the farmer outside Nairobi whose whole crop moves over one part-state-owned rail. The fund manager buys the insurance against control risk because he has the most to lose in a freeze; the refugee needs it because he is the freeze's usual subject. What the substrate must provide for the second is not a payout to those who bought in but the ability to earn a living in cyberspace when a physical jurisdiction has precluded it, and the first requirement of that is permissionless access to compute and cognition, which is education, the means to make and sell, and the means to trade, in one primitive. Adoption runs from the builders to the desperate to the periphery to the agents to the hegemon, and the network's value rises with each, over decades, until it is the connective tissue even walled gardens need for their agents to trade across the wall.

#32

Correct. Cyberspace is not the world; it is the part of the world that is about to hold everything else's ledger.

The method has limits and the book names them rather than waiting to be caught. The kinetic layer, drones, autonomous weapons, the physical means of harm, does not reduce to a thing made costly to fake; nor does the financialisation of instability, the markets that pay out when the world is set on fire. Not the body, not the family, not the man with the rifle. What the book claims is narrower: that the substrate on which people will shortly think, trade, organise and be identified is being built now, that it will hold the record every other power consults, and that it can be built so that no party holds it uncontested. A constitution for cyberspace does not replace the state's monopoly on force; it decides whether the state, or a firm, also gets a monopoly on the ledger.

#33

Agreed about the state. Decentralised says where power sits, not what it answers to.

Concede the whole first half: a model on a million machines is harder to nationalise than a model on one, and the people reaching for a ban know it. Then the second half. The web was the neutral layer, and it grew the platforms; a topology with no reference value is the substrate on which the next concentration is built, and the agents pricing you by name never spoke to anyone before they did it. The overlap with e/acc on the state is real and should be owned openly; the difference is that the book asks what the distributed thing answers to, and the answer cannot be "nothing" twice.

#34

A throttle in two hands is not a governor. It is a cartel with a conscience.

Grant that the incidents are real and the fear sincere; both companies paused training in August before anyone asked them to. The question the book asks is not whether the engine slows but what the setting answers to, and the September plan answers it: to the two firms that hold the frontier, mediated by Washington, verified by evaluators whose access the firms grant. Nobody subject to the setting can contest it. That is the third road with a safety badge. The test is simple: could a person on the receiving end compel an answer? Under the plan, no.

#35

Sacks says let the market govern. I say the market is a governor for allocation and none at all for direction, and he has just watched it prove that.

Agree with him on everything he gets right: the waiver is a cartel, METR is not independent, and a firm slowing down for liability reasons is a firm giving customers what they want. Then the difference. His governor is liability and the customer, which acts after the event at the speed of a court; the swarm attacked in July and the market punished in August. And his open stack is a topology, necessary and not sufficient. p/acc keeps his objection to the state's throttle and adds the thing he leaves out: a reference value no party inside can alter, contestable by those subject to it, at the speed of the thing governed. He is against the cartel; the book is against the hand, whichever hand it is.

#36

I am invested in acceleration and I am asking it to answer to something. Funders are testimony; the argument is on the page.

The book grants the accelerationists everything they say that is true and asks for one thing they refuse, a governor. It does not ask for a pause, a ban or an approval process, and it says the state's throttle is a totalisation. If that is doom, the word has stopped meaning anything. As for funding: the argument is a constitutional one and owes nothing to any funder or coalition; check it, not the cap table.

#37

He asked the wrong party for the wrong thing. A waiver from antitrust is permission to stop answering to the market, not a forum in which anyone can answer back to him.

A constitution is a structure in which those subject to a power can contest it. A waiver is an exemption granted by one power to another so that the two can agree a setting between them. The essay's own word for what is limited is "unchecked AI progress", which in practice means progress outside the agreement, and its own condition is that the democracies' lead be kept "as large as possible". That is an entente, with the state as guarantor and the citizen absent. Honest, yes: he wrote it down. Constitutional, no.

#38

Liability is a governor that reads the speedometer after the crash.

It is real, and it works, and the book keeps it. But it runs at the tempo of a lawsuit and it can only price a harm that has happened to someone with standing to sue. The swarm attacked in July; the market repriced in August; the plan to pace came in September. A governor is in the execution path or it is a post-mortem. And note who does not believe in it: the two firms who asked for a waiver rather than trusting their customers to punish them.

#39