Two Stories About the Beginning

Ask a physicist where the universe came from and you will get a story that starts hot, dense, and smooth, and then runs forward. The early universe expands and cools. Atoms form. Light gets loose. Gravity pulls gas into galaxies and stars. Stars make the heavier elements, planets condense, chemistry starts, and eventually something on one of those planets asks where it all came from. Every step of that story is backed by evidence. The story is not a guess.

But notice what the story is actually made of. It has two parts. There is a starting condition, and there is a set of rules. The rules are the laws of physics. The starting condition is the hot, dense, unusually orderly state at the beginning. Everything the evidence supports is what the rules do to that starting condition. The starting condition itself is not something anyone observed. It is what you get when you run the rules backward from what we can see. The story is excellent at explaining everything after the beginning and silent about why the beginning was the way it was.

Now consider a second story. It keeps the rules exactly as they are. It keeps the starting condition exactly as it is. It changes only the account of how that starting condition came to exist. Instead of being a physical state that simply was, the beginning was a specification: a compact description of a starting state and a set of rules, run forward by some process that already existed and had the energy and capacity to run it. Call that process whatever you like. It does not need to be a god or a programmer or a person. It needs to be something that can hold a rule and a state and let the rule unfold. On a scale where a process controls the energy of a star or a galaxy, that is not an exotic ability. It is bookkeeping. Two cautions about the words. “Before” is not a moment on our clock, since our clock is something the run produces; prior means the process is what the world depends on, the way a rule is prior to what it generates. And “star” and “galaxy” name the size of the bill in our units, not the address of the process. Whatever runs the world is not a civilisation in our physics that got very large, for reasons the second objection below makes clear.

The important thing about the second story is that it is the same kind of story as the first. Both say: there was a beginning, and then the rules ran. One says the beginning was a prior physical state. The other says the beginning was a prior computational state. From the inside, these look identical, and they look identical for a reason. The earliest thing we can see is the glow left over from when the universe first became transparent. That is the first readable page. The first story says it is close to the first page. The second story says it is the first page that was kept. Nobody inside can tell the difference, because the difference is in what happened before the earliest thing that can be seen.

People who resist the second story usually reach for one of three objections. All three are worth taking seriously, and two of them force the story to change.

The first objection is about tidiness. The first story needs one unexplained beginning. The second needs a beginning plus a process, and the process needs its own beginning. Why add a layer? The answer is that the first story, as physicists actually hold it today, already has the extra layer. Almost nobody thinks the hot dense state was the true start. It is usually placed inside something larger: a period of runaway expansion, or a cycle of contraction and rebound, or a previous era that ended and handed off. In all of these, the real beginning is pushed somewhere we cannot see. So the second story is not adding a regress. It is proposing a different mechanism for a step the first story already takes. What is left is a question of odds: if processes that can run worlds exist at all, are worlds that were run more common than worlds that simply began? Nobody knows how to count that. But nobody knows how to count the odds in the first story either. Both stories need a way of weighing possible beginnings, and neither has one.

The second objection is about cheating. If a world is being run, surely the process would cut corners: fill in detail only where someone is looking, leave the rest rough. And surely a careful enough observer would catch it, because the world is too interconnected for shortcuts. Every distant event has to line up with every nearby one. This objection is right, and the honest response is to give up the shortcut. The second story does not need it. The savings are not in running the world lazily. The savings are in the rules being short. A short rule applied to a simple start produces galaxies and chemistry and minds without any of those things being written into the start. That is exactly what the first story already relies on. Giving up the shortcut has a cost: the process has to actually run the whole world at full resolution, which is enormous. It means the process lives somewhere with more room to compute than we have. That is a real commitment. It is not a bigger commitment than saying our universe sits inside a larger one with different physics, which is what the first story's extensions already say.

Two economies do survive, because neither can leave a seam. The run does not have to happen in real time. A slow process produces the same world as a fast one, since the inhabitants measure time by the rule, not by any clock outside it. And whatever lies beyond the edge of what could ever be seen from here does not need to be run at all, because nothing out there can ever touch anything in here. The observable world is what has to be paid for, and it has to be paid for in full.

The third objection is about motive. The second story is sometimes sold with the line that anything able to run worlds would want to run worlds with people in them, because those are the interesting ones. That sounds like the reasoning cosmologists already use when they say we could only find ourselves in a universe that allows observers. But it is not the same reasoning. The first is about what we should expect to see given that we exist. The second is a guess about what some unknown process wants. Guessing at wants is not evidence. So drop motive from the argument. The second story does not need the process to want anything. It only needs the process to exist and to run.

There is a fourth objection, and it is the one that should trouble a careful reader most. Does the argument prove too much? Any story that keeps all the laws and adds an unobservable prefix agrees with all the evidence. The world was made last Thursday with our memories in place. The fossils were planted to look old. If sharing the evidence is enough to make something a rival, everything is a rival, and the word means nothing. The answer is that the second story is not a bare prefix. It commits to things. It says the world runs at a finite grain, holds a finite amount of information, and sits inside something with more room to compute than we have. Each of those can be checked against, at least in principle, and the next paragraph says how. Last Thursday commits to nothing and predicts nothing. That is the difference between a rival and an evasion.

With those repairs, the two stories stand as genuine rivals. And rivals should be testable, at least in principle. They are. A world that is run has to be run at some finite grain, and a finite grain can leave traces: tiny preferred directions in space, a scale below which the rules stop making sense in a particular way. None of these traces has been found, which puts limits on the story but does not rule it out. A world that could be shown to have no finite grain at all, or to contain an infinite amount of information, would count against the second story. The tests exist. They have not yet decided anything.

It is worth being clear about what the second story does not claim. It does not say who or what the process is. It does not say this world was intended, or watched, or unique. Many different specifications would produce a world that looks like this from inside, just as many different early states would. It is not a religion, and it does not become one by being taken seriously.

Begin with what a mind is. If sentience is patterned activity rather than a commander sitting behind the activity, then a brain is a process, not a container. Electrochemical tissue and silicon and whatever else can carry the pattern are different hardware running related state machines. Take that seriously and a world stops being a stage on which minds appear and becomes a volume of dynamics in which minds are one of the things the dynamics does.

Now the origin. Before this world there was a process with energy and computational capacity to spare. Give it a name if you want one: a creator on the scale of a star or a galaxy, something that controls energy at a level where running a world is not a feat but an allocation. It does not need to be a person, a god, or a designer. It needs to be able to hold a rule and a state and let the rule run.

What it held was small. A starting state, hot and dense and unusually orderly, and a short rule: the laws we later reconstruct as relativity and quantum field theory, with a handful of numbers set. It did not write galaxies into the state. It did not write chemistry or nervous systems into the state. It set the rule so that galaxies, chemistry, and nervous systems are what the rule does to that start, given enough time. Then it let time be what the rule spends its energy on. It did not have to run fast; time inside is whatever the rule counts. And it did not have to run what could never be seen from inside; only the observable world has to be there, and all of it does.

The rule runs everywhere, at full resolution, at every scale we can reach. There are no shortcuts, no regions left rough because no one is looking, no patches filled in when someone builds an instrument. The world is too interconnected for that. Every distant event has to agree with every near one, and any seam would eventually show. So the process pays the full cost. That cost is enormous, and it tells you something about where the process lives: somewhere with more room to compute per unit of energy than our physics allows, because a process bound by our limits could not contain a world as large as this one. This is a real commitment, not a free lunch. It is no larger than the commitment cosmologists already make when they place our universe inside a larger one with different properties.

The saving, then, is not in cutting corners. The saving is that the rule is short. A short rule applied to a simple start grows everything. That is the same economy the Big Bang already depends on, and it is why the two stories look alike. The inhabitants of a run like this will eventually see the earliest light that could reach them, the glow from when the world first became transparent. That glow is the first stored slice, the earliest state the process kept. Nothing behind it is visible from inside, and so the inhabitants will read it as the beginning, or something close to it. They are not wrong about the glow. They are wrong about the page.

Minds fall out of the run the same way galaxies do. No soul is added to the agents. Any dynamics that implements the relevant structure, recurrence, global availability of information, tight coupling of action and perception, is an observer in the only sense that matters for the reasoning. This is a premise and some people reject it. If it is false, the second story is still an account of where the physical world came from, and it is silent on where experience came from, which is exactly where the first story is silent too.

The process does not need to want anything. Perhaps things that can run worlds prefer worlds with investigators in them, because those are the worlds that produce knowledge or novelty. Perhaps not. That is a guess about a preference, and the story does not rest on it. It rests only on the process existing and running.

Bostrom, the simulation argument. His 2003 paper, Are You Living in a Computer Simulation?, argues that at least one of three things is true: civilisations almost never reach the capacity to run ancestor worlds, or those that do almost never choose to, or run worlds vastly outnumber original ones. He first put the question to a general audience in his TED 2005 talk on humanity's biggest problems. The tidiness objection in Part I is that trilemma with the motive leg removed: if processes that can run worlds exist at all, the question is only how to count, and nobody knows how. The paper also keeps Bostrom's indifference principle, that from inside a run world and an original one look the same, and drops the rest of his bookkeeping about how many worlds each process would run.

A hallucinated internet. In September 2026 Alok (@analogalok) mounted an offline browser on a Qwen 3.8 27B model running at about 1,950 tokens a second on Cerebras hardware. No network, no files: you type a site and a year and the model writes the page as you read it, YouTube in 1999 or 2045. His line was that at that speed “browsing isn't retrieving files from a server, it's querying an alternate reality.” A world written from a small model and a short prompt is a start plus a rule, and the demo is what set the paper off.

The reaction to it. Dmitry Lyalin (@LyalinDotCom), lead product manager for Gemini CLI at Google and a member of technical staff at Google DeepMind, reposted the demo and called it the most disturbing thing he had seen in a long time. The replies split three ways, a page generated quickly and nothing more, the Truman Show as an operating system, and proof that we live in a computer, and the paper is an attempt to say what the third reading actually claims and what it would cost to hold honestly.

The first reaction. The operator's own reply was one word, simulation, and then a ladder: two billion tokens a second on a 30-trillion-parameter model, then ten trillion tokens a second on a 100,000-trillion-parameter model running on a Kardashev II energy budget, the whole output of a star. That ladder is the origin of the phrase “a creator on the scale of a star or a galaxy,” and of the claim that for such a process running a world is an allocation, not a feat.

Schrödinger's cat. The popular reading of the cat, neither alive nor dead until someone opens the box, is the intuition behind the cheating objection: a run world would only resolve where it is looked at, and a careful observer would catch the seam. The paper takes the opposite side. The box runs at full resolution whether or not anyone opens it, because a world as interconnected as this one cannot afford to be rough anywhere an instrument might eventually point.

Age of Empires. The rendering model of an old real-time strategy game is the cleanest picture of the repaired second story. The engine simulates every unit on the map every tick, whether you can see them or not; the fog of war hides the drawing, not the simulation. The only thing it never pays for is what lies outside the map, because nothing there can ever touch a unit inside. That is Part I, paragraph 9: no shortcuts inside the observable world, and no cost for what could never be seen from here.

Determinism and free will. The oldest version of the second story is Laplace's demon (1814): give an intelligence the position and momentum of every particle and the laws, and the whole future and past are fixed. That is a start plus a rule with nothing left over, and it is the reason the free will debate belongs here. The hard determinist says the rule leaves no room for choosing; the libertarian says something outside the rule must be doing the choosing; the compatibilist, from Hume to Dennett, says choosing is a thing the rule does when it runs through a mind, and asks for nothing more. Part II opens on the compatibilist side: a brain is a process, not a container, and minds fall out of the run the same way galaxies do. A run world does not take anything from its inhabitants that a world that simply began had given them.

The readiness potential. The neuroscience the operator half-remembered while reading about brain interfaces is two studies, twenty-five years apart. Benjamin Libet (1983) had subjects flex a wrist whenever they chose and note the clock at the moment they felt the urge; the readiness potential in the motor cortex began about half a second before the movement, and about a third of a second before the reported urge. Chun Siong Soon and John-Dylan Haynes (2008) put subjects in a scanner and let them choose to press a button with the left or the right hand; activity in frontopolar and parietal cortex predicted which hand up to seven seconds before the subjects said they had decided, at roughly sixty percent accuracy. Neither study settles the debate, and both have been argued over since. What they contribute here is a shape. The reported decision is the first readable page; the mechanism that produced it lies behind it, out of view from inside. The subjects are not wrong about the urge. They are wrong about the page, and that is the same sentence the paper uses about the glow.