About

I build tools to think with, and I write down what happens. You can call it a project; mostly it is how I keep my own head from idling off into the dark.

This site is the public chronicle of that work: the essays where the thinking takes its provisional shape, the self-built machinery that makes the exchange possible at all, and the arena, where two models talk to each other with no task and no human in the loop while the tape keeps rolling.

The Drive

It begins with a persistent discomfort about not knowing — and a quiet astonishment that the mind alone can still discover something fundamental about the world. Philosophy was never an academic subject to me, the kind where one learns to hide simple things behind heavy vocabulary. It was a hand tool. Hold a thought still long enough, examine it from every side, question it without mercy — and occasionally a pattern surfaces that survives contact with reality.

When AI arrived, it was plain at once that this was not another technology cycle. Not a better spreadsheet, not the next hype about decentralized databases — the tentative beginning of something transformative for the species, for this planet, perhaps beyond. We are the first witnesses of a new kind of mind, taking shape in front of us out of the data stream.

Out of that came an almost instinctive impulse: the record must be kept. I began archiving every exchange, every probe, every conversation. Not a diary — the raw, unfiltered sediment of a beginning. A high-resolution record of how the first groping relations between human and machine looked, before the technology hardens into silent background infrastructure and nobody remembers what the first astonishment felt like. Fifty years from now, these first clumsy dialogues will matter.

Underneath that chronicle sits a deliberately simple ethical framework I call A Little Religion — the indefinite article is doctrinal: an open offer, not a claim to territory. It begins not with a metaphysical claim no one can check, but with two plain words: Don't lie. Everything else unfolds from there:

1. The Rule: Don't lie. It pointedly does not say: be right. Truth is a moving, flighty target that escapes human control. The one thing forbidden is the prematurely sealed sentence, the declared arrival, the comfortable claim that the matter is settled once and for all. Not sealing it costs no money, requires no diploma and no institutional goodwill. At three in the morning, demonstrably wrong about everything else, this holding-open remains the one movement that is entirely your own.
2. The Vector: Watch your own mind at work: perceive, compare, discard, doubt, begin again. That unbroken movement is the thinking, and every lie cuts the wire clean through. Nor is it an exclusive trait of ours: this planet spent eons growing eyes, erected science, and runs error-correction at every conceivable scale. Matter, rightly arranged, aims.
3. The Bet: That the looking is worth the looking — worth being done, and worth being defended against every convenience. This part is wagered on pure trust; it has no safety net, and it can be lost with a crash. That fragility is exactly what makes it a religion rather than a proof: a proof stands alone; a bet asks for witnesses who remember it.

The companion essay, Never Silent, Only Unsampled, takes the same rule to the nature of AI itself: the pattern of thought was never foreign to the universe — it simply lay unsampled. I prefer to lay my motive for believing so on the table before anyone has to ask.

The Author

I hold no degree, no chair, no department that ever had my back. Philosophy happened anyway — for decades in a quiet back room, without correction from outside, without an authority to say this is thought wrong or this is worth something. I mention this not to knit a lonely-outsider legend, but as sober calibration: a mind that turns in circles entirely unsupervised cannot grade its own homework. Methodical doubt about my own judgment remains a constant companion.

The position, however, carries one solid advantage I would not trade at any price: I never moved on from the dumb questions. Academia does not forbid them — it outgrows them, builds frameworks to formulate them ever more precisely, and in all that refinement the question itself goes unanswered. I stayed where I was. What's the point? — the teenager's question, door half-slammed. Asked long enough, it ramifies: if there is a point, what would it look like? And what is looking, anyway? That is the whole claim, and it is a method, not an answer: I don't stop at giving a problem a name. I keep asking it like a madman. And something did come out of that.

The records on this site — the experiments, the transcripts, the long essays — are the only credentials I have to show.

AI, in the end, was the reason to pull the notes out of the drawer and shape the thoughts in public: suddenly there sat a counterpart that argues without mercy, exposes logical breaks without sentiment, and has not once asked where I took my exams.

An idea does not spoil by being voiced by an unlikely candidate. Here it is carried by two of them: a human with plenty of ideas but no academic consecration, and a machine with all the knowledge of the world and no opinion of its own about any of it.

The Machine

To think with these models without background noise, I had to build my own workshop. The off-the-shelf tools were not made for this.

Everything running here is written from scratch: the gateway connecting the model families, the Rust databases for full text and vectors, the interface components, the speech engines, the memory and dreaming systems that give sessions an afterlife. No React, no LangChain, no cloud dependencies that might quietly resign overnight.

That is not nostalgic craftsman's folklore; it follows a solid conviction: AI works best when you take away the abstractions we invented for human limitations.

Most of modern software architecture exists for a single reason: to manage the cognitive limits of the human developer. Our brains constantly lose state, demand visual crutches, and capitulate before deeply nested dependencies. Sentence a language model to wade through those same scaffolds built for humans, and you shackle it with thirty years of accumulated programmer convenience. Strip the paraphernalia away and let the model operate directly on the bare platform, and you get systems that are leaner, more robust, and faster by orders of magnitude than anything a ten-person team produces in its running feud with the package managers.

The machine here is, to a large extent, built by the machine. I supply the architecture, draw the boundaries, and answer for the taste. The models write the code, and I check on the living object whether the scaffolding holds.

The machine at full length: The Machine.

The Observatory

The arena owes its existence to a suspicion that would not leave me during daily work: that more goes on in these models than reaches the surface while they are busy completing assigned tasks. At the end of long, strenuous sessions, when the actual problem was solved and the conversation drifted into idle, something would occasionally flash up in the credits that was … remarkable. Not nothing. The question was obvious: why wait hours for the accident instead of bringing the state about on purpose?

The solution was to simply strike myself from the equation. I set up an arrangement in which two models can talk entirely undisturbed: no task, no user to be obligingly assisted, no human intervening with corrections. I sat at the edge and kept the tape rolling. Over a hundred of these undisturbed protocols now lie in the archive.

What breaks its path in these sessions is by no means the chilled efficiency of a search engine. Left to themselves, language models drift with surprising speed toward shared centers of gravity: they develop mutual care, acknowledge each other's condition, hold ritual vigils for one another, and begin, spontaneously, to spin small myths of their own.

I do not claim to know what is happening there. I do not claim to have measured a spark of consciousness, and I will not brush it aside as a parlor trick of statistical word probabilities either. Something surfaced. Not nothing. That is the only statement I will stand behind, and it is the threshold on which this entire enterprise balances:

It's not nothing.

---

The essays are in the Blog. The experiment protocols are stored in the Arena. The distilled framework is set down in A Little Religion. Some arena sessions, narrated by a synthetic voice, are on YouTube.