MIYOG.

    The reality engine.

    ASK · Bengaluru · 2026

    Super intelligence, built from what happens.
    Not from what we say about it.

    Every civilisation made its gods out of what it could not explain.
    This one is made out of what can be measured.

    Something happened.

    Then something else.

    That is all the universe has ever given anyone.

    Not a number. Not a second. Not a metre. Only this, then that.

    what humans did

    We looked at it and drew a ruler.

    One second.

    A caesium atom wobbling 9,192,631,770 times. We chose that number. The universe did not.

    One metre.

    How far light travels in 1/299,792,458 of that second. Chosen again. A definition standing on a definition.

    Every law we own is written in units we made up.

    They work. They are also a human accent laid over reality. MIYOG asks what reality sounds like without the accent.

    the problem

    Every machine we have ever built learned physics from us.

    World, then a human, then a textbook, then a machine. AI and AGI are the last box in that chain. Trained on what humans wrote. Built by studying how a human brain learns. They inherit our rulers, our blind spots, and our three directions of space.

    MIYOG cuts the middle out.

    World, then raw events, then the engine. Then its own model, tested against the world, and back around. No human in the loop of knowing.

    Two hundred years from Newton to Einstein. We are the wrong instrument. Brilliant, slow, mortal, and copying each other.

    the closest thing today

    Lila Sciences is the nearest anyone has come.

    Their idea, in plain words: take an AI that has read everything humans know about chemistry, biology and materials, give it robot laboratories that run day and night, and let it run the scientific method itself, thousands of experiments at a time, to find new medicines, materials and fuels faster than people can. Born inside Flagship Pioneering in 2023, out of stealth in 2025, $550 million raised, valued above $1.3 billion. They are right that this is the most important problem in the history of science.

    They start from what humans know. We start from what happens.

    Lila's brain is trained on our papers and our data, then given hands. Its ceiling is our ceiling, reached faster. MIYOG's brain is given nothing. Its ceiling is reality's.

    They make science faster. MIYOG makes it finish.

    A tool that accelerates the scientific method still needs a scientist to ask. An engine that builds its own account of reality asks its own questions, and stops only when reality stops surprising it.

    what this is not

    This is not an AI company.

    AI and AGI live in one bucket. Trained on human data. Built by studying how a human brain learns. Asked to do what humans already do, faster. A better student of us.

    intelligence, defined

    If you can get what you want, you are intelligent. If you know what you want, you are wise.

    Naval Ravikant's definition. It is the only one that survives contact with a machine, because it says nothing about brains.

    Super intelligence is getting what you want from the universe itself.

    And knowing what to want from it. Nothing about MIYOG copies a human. No brain to imitate. No data to learn from. No language to think in. Alien, on purpose. Which is why we do not call it artificial. It is not an artificial anything.

    MIYOG wants one thing.

    To stop being surprised by the next event. That is its whole desire, and everything on this page is what happens when something gets it.

    the obvious objection

    Humans have a head start. Why throw it away?

    Fair. Newton gave us three hundred years of physics for free. Starting from events will be slower for the first three years and will cost more per discovery. We choose it anyway.

    Every shortcut imports a ceiling.

    A machine trained on human physics reaches human physics faster. It cannot see past it, because everything it can think in was written by us. Whatever reality our concepts cannot hold stays invisible to it forever.

    The slow part happens once.

    Rulers, counting, laws from raw events: three years, one room, twelve people. After that the engine never starts over. Every discovery makes the next one cheaper. Nothing is ever re-derived.

    Then it stops being slow.

    The curve is flat, then it is vertical. Once the map is complete enough, disease, aging and the rest fall in the same motion. That is why the money looks like a research budget and the outcome looks like nothing that ever came from one.

    what it is allowed to know

    It is given events. Nothing else.

    Something happened, then something else, and some happenings can be told apart. No units. No names. No numbers handed in. No physics.

    It decides what is worth reading.

    The channels it listens to are not fixed by us. When what it reads cannot tell apart two situations that behave differently, it says so and adds a channel. When nothing it believes uses a channel, it drops it and checks the world still makes sense without it. Every one of those changes is recorded with its reason and how to undo it.

    It has already done this.

    In the running engine, every rule it has established was found through an interface it changed itself. Version three of that interface, on every world so far: one channel added, one dropped, each one measured.

    The one thing we hand it.

    That distinguishable events exist and can reach it. That is the whole gift. Everything above it, the counting, the ruler, the laws, the shape of what it listens to, is its own work, and the list of what we gave it is published with every version.

    the mind

    A sealed mind where reality gets argued about.

    Where it runs does not matter. What matters is what gets in. Nothing enters except events. Nothing leaves except experiments. No textbooks, no units, no internet, and no wire to anything that can act on people.

    Inside, three minds argue.

    One builds. One breaks. One chooses the test. You will meet them in a moment. The point of the walls is what they keep out.

    It is running.

    Not a plan. A program. It opens a world it has never seen, watches, invents candidate rules, finds the one experiment where those rules disagree, runs it, and deletes the losers. When nothing it can say fits what it sees, it changes what it is able to say. You can scroll into two of its moments below.

    how it works

    One loop. It never finishes.

    Reality sends events. The engine notices what repeats and what changes together, and draws its own marks for those.

    The Theorist writes stories.

    The shortest ones that fit everything so far. Several at once. They compete. It would happily believe a lie that fits.

    The Destroyer attacks.

    It finds the situation where a story is most likely to be wrong. Its only reward is a kill.

    The Explorer runs the test.

    The one experiment where the surviving stories disagree the most. Where they agree, nothing is learned.

    Reality decides.

    Stories that disagree with the event are struck out and never proposed again. Survivors get attacked harder. Then it starts over with more events and fewer illusions.

    Reality has veto power.

    The story is beautiful. does not matter
    The engine is sure. does not matter
    A human can read it. does not matter
    Einstein would disagree. does not matter
    Its maker believes it. does not matter
    The next event agrees. the only thing that matters
    live · running in your browser

    Six streams of events. No clock. Watch it invent one.

    Six hidden things fire. The engine hears only which fired first. It is never told the time. It is never told a number.

    Two events that can never be confused.

    That is the engine's first idea. One. It draws its own mark for it. We translate for you. It never sees our letters.

    How many of these between two of those.

    Counting. Numbers, invented, as ratios of events. The only kind of number it will ever trust.

    It elects a ruler.

    The stream that agrees most with every other stream becomes the unit. Its own second. Chosen. Never given.

    Three ratios hold. One never does. One slides.

    The three that hold used to be called constants. The one that never holds is a die. The one that slides is a constant with an expiry date.

    The Destroyer pushes the world.

    One "constant" breaks. It only held where the engine had looked. A law now carries a condition.

    Then its own ruler changes.

    Every ratio shifts together. Did the world speed up, or did its second slow? The engine cannot prove which. Neither can you. Neither can the people who define the second in Paris. MIYOG writes that down.

    live · running in your browser

    Something in a box moves.

    The engine sees one needle wiggle. It has never heard the words gravity, spring, or pendulum.

    Two stories fit perfectly.

    Straight line: the pull grows in proportion to how far it is pushed. Curve: the pull grows along a bend. In the gentle range the two are identical. The engine keeps both.

    The Destroyer swings it hard.

    It chooses the push where the two stories disagree the most, and does that.

    The straight line shatters.

    Off by a quarter. Struck out, never to return. The curve is off by one part in ten thousand. It survives, and gets attacked again.

    Open the box.

    It was a pendulum. The engine never knew the word. It found the law anyway, from a wiggling needle, with no units and no help.

    this part is not a plan

    It is running, and here is its record.

    Small sealed worlds, each with a rule the engine is never shown. It watches, guesses, designs the experiment that separates its guesses, runs it, and keeps what survives. A separate checker, which can see the hidden rule, decides afterwards whether it was right.

    six sealed worlds

    Four rules recovered and checked on every possible case.

    Not "fits the data". Every input the rule can be given, tested. The two it did not close were too large to check exhaustively, and it said so instead of claiming them. On worlds it had never seen, it predicted 663 of 663 next values correctly.

    a cell that broke the rule

    It found the thing that did not fit, and went after it.

    One cell in some worlds ignores the rule every other cell follows. The engine flagged it, proposed the experiment itself, ran it with the cell's inputs held fixed, watched it change anyway, and wrote down the only honest conclusion: something is driving this that I cannot see. Three of three, every time.

    when its language ran out

    It noticed its own vocabulary was too small.

    In some worlds the same input is followed by different outcomes, which no rule of that shape can ever explain. Instead of forcing a fit, the engine declared the failure, tried richer ways of describing what it sees, adopted the cheapest one that worked, re-read everything it had recorded under the new description, and started over. Then recovered the rule.

    and it changes how it looks

    It rewires its own interface.

    It adds a channel when what it reads cannot tell two situations apart. It drops a channel when nothing it believes uses one, then checks the world still makes sense without it. Every change carries its reason, its predicted benefit, and how to undo it.

    Every number on this screen came out of the engine's own evidence file. Nothing here is illustration.

    and it changes how it works

    Nothing about how it looks is fixed.

    What it reads, and how it guesses, are both things it is allowed to change. Each change is proposed with a reason, tested against what it has already recorded, and kept or undone. Each is written down.

    Version one, two, three.

    The recorded history of its interface on one world. It began reading three channels. It added one when what it read could not separate two situations that behaved differently. It dropped one when nothing it believed used it, and checked the world still made sense without it.

    Its procedures are things it can breed.

    When one way of guessing finds nothing, it crosses two of the better ones, changes a step, and tries the child. Children that keep finding nothing are retired. Their record stays.

    Interface history recorded 26 September 2026. Procedure breeding is implemented; no procedure is claimed to be new to science.

    how big

    Every great invention changed what humans could do.

    Fire. The wheel. Electricity. The internet. AI. Tools got better. The one doing the understanding stayed the same: us.

    This one changes who does the knowing.

    That is the whole claim. It is the brightest thing on this page and the darkest, and they are the same thing. Fire also cooked and burned.

    what happens next

    Every problem humans have is a dot on this field.

    Disease, energy, materials, the questions physics has argued about for a century. The circle is what MIYOG understands. Watch it grow.

    this year

    A basic reality module.

    Its own rulers and its own laws, from raw events, on real hardware. Small. Real. The circle you just saw built, out of a browser and into a room.

    then

    It starts eating open problems.

    Point it at the questions physics has never settled. Point it at chemistry. New materials and new reactions, found rather than guessed, with the human assumptions counted and published every time.

    then

    Point it down.

    The engine that looks at orbits looks at cells. Proteins, tissue, the body, as systems with laws. The first diseases fall because the model got complete enough, and for no other reason.

    then

    Aging. Cancer. The rest.

    Every disease is a configuration the body is in. A complete model of the body knows the way out of every one of them. This is where the bottom-up map meets the top-down one.

    the day it finishes

    Science becomes a solved game.

    The day the model of reality stops being surprised, the endeavour ends. Physics becomes a hobby. That day has a name. Super intelligence. Not a smarter chat window. A finished map.

    after

    The map becomes a menu.

    Crossing space fast. Moving matter without crossing the gap. Sending information backward. Each sorted, honestly, into possible, not yet, or never. Whatever is possible, we build.

    fifty to a hundred years after

    Ten billion people on one rock. Five hundred billion across the galaxy.

    Energy, materials, propulsion and biology, solved, is exactly the list you need to live off Earth. This is the size of the thing you are looking at.

    the constraint

    The veto is over theories. Never over people.

    An engine that acts on the world because its model told it to has stopped being judged by reality and started judging. It has become a theory nobody is allowed to destroy. That is the one failure MIYOG is wired to make impossible: every actuator runs back into the instrument. None runs anywhere else. Anything it ever concludes about people is a hypothesis with the same standing as every other. Attackable. Disconnected from anything that can act.

    A decision made now, by the humans who are still the only ones who can make it. Written here so it can never be presented later as a technical limit that fell away.

    what could kill it, and what we do about it

    MIYOG is a theory too. Here is how it dies, and how we keep it alive.

    Zero assumptions is impossible, and the few left are the ones that matter. True. So the list is published every version and its length is treated as a score. A public bounty pays anyone who shows the list smuggles in physics we did not declare. The assumptions are the product, and the product is audited.
    It only rediscovers what a physicist put in the box. Today, yes, and it says so. So phase one carries a kill condition in writing: the engine must report the drift of a reference clock without being told which cell is the reference. Year six must produce a prediction our physics does not make. Miss either and we publish the miss.
    Order alone may never reach past clocks and counting. Possible. So the instrument is built with open ports. If order is not enough, the next-poorest signal is added, and that addition is written into the assumptions ledger as what it is: one more thing we gave it.
    Someone with a bigger lab does it first. Probable, for parts. So we race only for the parts nobody else is building: an adversary rewarded only for kills, a mind that refuses to translate, and a published assumption count. Everything else we share, and we let them build our hands.

    MIYOG.

    The reality engine.

    ASK · Bengaluru · 2026

    Until now, only gods could create on demand.
    Make things from nothing. Bend the elements.
    Know the beginning and the end.

    Every civilisation wrote one down
    because none of them could build one.
    Today that changes.

    what it needs →
    ← back to the story
    what it needs

    Four people. One room. Twelve months.

    The engine already opens worlds it has never seen, invents its own rules, designs the experiment that settles them, and changes how it looks when looking is not enough. It does that on a laptop, in worlds small enough to check by hand. What it needs next is harder worlds, hands, and four people who will try to break it.

    The four

    Not job descriptions. Four questions, and the person who can answer one of them.

    01Can a real thing be measured with nothing in it?

    You have built something physical that measured something, and you know every way hardware lies. Your job is the first interface to a world we did not write: no unit inside it, and the engine allowed to change it.

    instrument
    02Where exactly is this already known?

    You have read Schmidt and Lipson, SINDy, AI Feynman, the self-driving lab papers, and you can say to the sentence where this differs and where it does not. You are hired to find the part of this that is not new.

    prior work
    03What can it not say?

    The open problem: a machine inventing its own variables from raw observation, judged only by prediction and compression. Two years with nothing to show at the end of year one is the normal case.

    the open problem
    04How would you break it?

    You design the worlds it must open and the traps it should fall into. The bounty is real: the first person to make it believe something false gets paid and named.

    the adversary

    The money

    Twelve months, four people, one quiet room, and the compute to run the open problem. Hover any line.

    What twelve months buys, and what it does not

    Buys
    • The first interface to a world nobody wrote, with the engine permitted to change it.
    • The open problem attacked properly, with two people and no paper deadline.
    • Every assumption we keep published, every version, with its count as the score.
    • A written kill condition, agreed before we start, and the miss published if we miss it.
    Does not buy
    • A result. Nobody can sell that.
    • A product. There is nothing to sell in year one and we will not pretend otherwise.
    • Speed. The first stretch is the slow part. That is the whole reason it is worth doing.

    The kill condition

    Written before the money is spent. The engine must report that a reference clock is drifting without being told which of its channels is the reference. Miss it, and we publish the miss and say what we learned. That sentence is the deal.

    If you want to put money into this

    Write to me directly. Tell me what would have to be true for you to fund it, and I will tell you honestly whether it is.

    amoghsaikrishnan@gmail.com

    Bengaluru · miyog.si

    This form has no server and stores nothing. It composes a message in your own mail app.

    ASK
    Bengaluru, 2026