At breakfast one morning in March I asked an innocent technical question, and by midnight the kitchen table was covered in diagrams and I understood something new about every word I have ever used.

The question was about AI, which I have been studying all year with the screwdriver, as usual. When a machine works with the meaning of a word, what is it actually holding? The answer, I knew vaguely, is a list of numbers, hundreds of them per word, called an embedding. Fine. But at breakfast the real question surfaced: where do the numbers come from? Who told the machine that king should sit near queen, that ocean should sit near wave and far from spreadsheet?

The answer is the single most elegant thing I have learned this year. Nobody told it. Nobody could have. The machine reads billions of sentences and plays one endless guessing game, predict the missing word, and every wrong guess nudges the numbers a little. That is the entire method. And out of nothing but that, out of pure exposure to what shows up near what, a geometry assembles itself. Words used in similar company drift together, like guests at a long party finding their people. King and queen end up neighbors. Storm settles near trouble in one dialect of text, and near surfing in another. Meaning, it turns out, is not stored anywhere as a definition. Meaning is a location, and the location was learned entirely from context.

By noon I had the diagrams out, because the implication had walked in and sat down at the table, and it concerned the other meaning-machine in the house.

Where did my meanings come from? Nobody handed a child a dictionary entry for love, or for quiet, or for himself. A child learns those words the exact same way, from context, from what showed up near them, thousands of times, in one particular house, one particular decade. If raised voices kept company with evenings, the geometry recorded the neighborhood. If praise sat near achievement and never near rest, that adjacency went in too, silently, structurally, in whatever tissue holds a person's map of what things mean. None of it chosen. All of it learned, flawlessly, from the local data. Your private definitions, the ones that steer your days from below, are not truths. They are geographies. Coordinates, assembled by exposure, in a space nobody ever showed you the map of.

And then the evening's discovery, the one that had me pacing the balcony in the March cold, grinning at the city. The machine's map is not finished. That is the point of it. Show the model new text, new company for old words, and the coordinates move, that is what learning is, the geometry never sets like concrete. And every honest thing I have logged in this notebook these four years says the same about ours. Quiet has been migrating in me, entry by entry, from its old neighborhood, suspicious, loading, alone, across the whole map, to where it lives now, next to sea, next to morning, next to home. Nobody relocated it by argument. It moved the way it was placed: by context. By two years of quiet keeping better company.

So this is where meanings live: in a space that experience built and experience can renovate. You cannot think your way to the new coordinates, I have tried, thinking checks the old map. You have to feed the words new company, day after ordinary day, and let the geometry do what geometry does.

Choose your words' neighbors carefully, in other words. You are, at all times, at breakfast and at midnight, someone's context. Most importantly, your own.