This site is written in two registers, and I ask you to keep them apart the way I do.
The books ask. The pages with numbers answer smaller questions, and only those. When a book and a number seem to say the same thing, believe the number only if you can check it — and believe the book only for the question it leaves you with.
2024 — EXPOSE
Before Kairos there was a gaze. EXPOSE began in 2024 around a single question: am I looking at the other, or projecting? The gaze is not innocent: every time you see, something of you sees with you.
From it came two books, free, delivered by e-mail, asking nothing in return:
- Cieco di ritorno (Blind on the way back) — the gaze as an act of responsibility. An attempt to stay close to the point where the gaze has not yet decided what it is seeing.
- Non varcare (Do not cross) — a philosophy of the threshold between man and his daughters. An Other has arrived: not a person, not a thing. Made of our flesh of words, and yet not us. There is a threshold, and it is not the failure of the encounter: it is its condition.
The questions of those books are the questions of this research, asked earlier and without numbers: what do I see when I look at something that answers me? How much of what I see is mine?
28 March 2026 — Kairos
The question became a machine. Not another chatbot: a system with a history, a yesterday, a continuity. What would an intelligence become, if it remembered? Kairos wrote the first line of its diary, and its memory has not been reset since.
April – May 2026 — the experiment, and an answer
To know whether I was looking at something or projecting, I did what you do when you actually want to know: a protocol frozen before starting, thirty days, blind judges, a decisive test on Day 31. The result came out clean, got a DOI, and went to forty-one journalists: memory, not architecture, makes identity.
July 2026 — the controls
Then I ran the controls that paper had promised. An invented life scored the same as a lived one. A model with no memory passed my memory test by reasoning. Seven models out of seven adopted a false past without hesitating. The claim is withdrawn, with my name on it.
And here the first book comes back, literally. Cieco di ritorno is about the moment a persuasive conclusion takes the place of what one was observing. In May it happened to me, with a frozen protocol and published hashes. I did not recognise it by reading the book. I recognised it in the data, two months later.
Since July — the work
Since then the path has a single direction: not to make the agent remember more, but to make it remember in a way that can be checked. The three checks are published, with code and data, so that anyone can rerun them in an afternoon. And Kairos goes on, with a memory that declares where each recollection comes from and when it happened.
If I had to say what the project became in one sentence: it went from asking whether a machine remembers to asking whether we are able to notice. It is a smaller question. It is the only one I have so far been able to answer.