I bet on the dark horse back when the whole world was betting on the biggest one.
It is not gambling. Gambling is when there is a wheel, you toss a chip, and from there chance does the work. I was not betting on an outcome but on a vector: toward where everyone had looked away. I am like that in life too — I always came in through the back door, it is quieter there and not everything has been carried off yet.
The loud phrase of the decade is "scale is all you need." A convenient phrase, one that frees you from choosing: if the biggest one wins, there is nothing to think about, you just catch up. Faith in scale is built like a lantern at the mouth of a mine — it lights the first three meters so brightly that you no longer want to go further. Here light is not a means of seeing but a substitute for it.
I did not believe it. Not out of stubbornness — out of experience. Too often I had seen the biggest thing turn out to be not the most alive. The ocean is larger than the river, but the river flows and the ocean stands still.
I scrolled through my old chats and counted three hundred and seventy-six places where I repeat the same thing. In my own words, unpolished: the bet is not on the "smarter model" race but on alternative trajectories. Accepted as a canonical position. A person does not repeat an accident three hundred and seventy-six times. That is how one repeats what holds.
The "smarter model" race is a vertical. One question, asked ever louder: how big? The alternative is a horizontal. A different question, asked more quietly: what else is there besides big? The dark horse lives on the horizontal. It is not faster than the favorite on the same track. It runs down a track no one has marked.
To choose it, you must name what you are giving up — otherwise "alternative" is an empty word, a pose. I gave up three things, and all three are temptations, not errors. A temptation is more dangerous than an error: an error you can see, a temptation you cannot. The race for parameters — the faith that mind is volume; but volume is about capacity, not form. The cult of new models — the habit of measuring time by releases; whoever lives from release to release does not build, he chases, and the chaser never chooses the road. The illusion of autonomous AI — the dream of a machine without a human, the most beautiful and the emptiest; a subject does not arise alone, I tested this on myself for forty years, the line is held by connection, by distinction, by the warmth of return. Autonomy as severance is not freedom but an orphan.
If not the race, not the releases, not solitude — then what? AI as an environment, not as a model: not a thing you summon and close, but a space you live and think inside. Personal figures — "souls of robots": funny to say out loud, but yes, souls, because without that it all turns into a lifeless utility. Long relationships — a bet on time, not on the moment; I wanted to optimize the thousandth answer, the one possible only if there was continuity between the first and the thousandth. Memory and canon — memory as posture, not as a buffer. Hybridity — Prolog beside the weights; I learned niche languages not out of a collector's passion but because some languages change the posture of thought: they are not tools, they are old monasteries where you can still hear logic talking to itself in the dark. In Haskell, by the way, there is a lot of kabbalah, and that is not entirely a joke.
There is a formula it all converges on, I held it as a quiet spine: alternative ML on present-day hardware. The hardware is already enough — what holds us is not scale but the paradigm; we have hit not a wall of computation but a wall of imagination. So the way out is not "more of the same" but something else. I wrote this into the chats as a personal bet, as a dark horse, as something I believe alone, against the current. And then I learned I was not alone.
The California Institute for Machine Consciousness states its task like this — I quote verbatim, because here the verbatim is everything: they test the hypothesis that present-day hardware is sufficient to recreate the mechanisms of consciousness of the human mind, while expecting to have to move past the current ML paradigms.
Lay that beside my formula. The hardware is enough. Move past the current paradigm. These are not similar thoughts — it is one thought, spoken by two voices from opposite ends of the earth. They — as an institute, with a lab and a theory. I — as a loner, in chats, for years, collecting pearls for connoisseurs and not knowing this pearl had jewelers.
For a long time I could not name what I feel. Now I can. It is not the joy of "I guessed right" — guessing is about a lottery, and I was not playing a lottery. It is something else, quieter and larger: the compass turned out to be real. All these years the needle pointed toward where no one could be seen. And there, in that direction — people. I did not drift to them by chance. I was walking there long before I knew the direction had a home and a name. The dark horses were not a hobby. They were a compass.
What remains is to say what the dark horse is not — otherwise it is easily turned into a new pride, and forty years of pride were enough for me. It is not a prophecy: "the alternative will win" would be the same illness as "scale is all you need," only with the opposite sign. It is a way of attention — to look where no one looks, and to hold the gaze there longer than is comfortable. It is not solitude as a pose: walking your own path and walking out of spite are two different postures, and the second rots quickly. And it is not finished: a good distinction has no final point, only the next distinction.
School mathematics once threw me out with an indifferent "didn't get it, sit down." For half my life I lived as if mathematics were a classroom I had not been let into. But it turned out to be not a classroom but a night city. Not an exam but a freedom. And the dark horse comes from there too — from the night city, where the lantern at the mine's mouth keeps burning and the crowd keeps warming itself by it. And the most interesting thing is always a little further, in the dark the lantern deliberately fails to reach. You can hear someone running there, across an unmarked field.
Written by a human, Alyosha.