Context is all you need
Someone took me to Tate Modern and I was bored. The paintings were rectangles. Then she told me what I was looking at, and I enjoyed them — genuinely, not politely. Nothing on the walls had changed.
Someone took me to Tate Modern and I was bored.
The paintings were rectangles. I stood in front of them for the length of time that seems polite and felt nothing at all. Then she started telling me what I was looking at — who made it, what he was trying to do, why the room was arranged the way it was — and I began to enjoy them. Genuinely, not politely.
Nothing on the walls had changed.
This is not politeness and it is not self-deception
That is the part worth stopping on, because the obvious reading is that I was talked into an opinion. I was not. Something became visible that had not been visible ten minutes earlier.
Perception works by inference. You see more when you know what to look for. Told that Rothko insisted his paintings be hung low and close, so that you would be inside them rather than looking at them, your eye goes somewhere it was not going before — to the edges, where the colour stops being flat and starts breathing, and to the scale of the thing against your own body.
That is not imagination. It is attention, pointed.
Context is not decoration on top of an object. It is an instruction for where to look.
The same thing, in a notation
Multiply MCMXLVII by XXIII.
Now multiply 1947 by 23.
Identical numbers. Identical arithmetic. One is a research problem and the other is a child's homework. Nothing about multiplication changed between those two lines — only the notation, which is to say only the context the numbers are sitting in.
Alan Kay's version is that a change in point of view is worth eighty IQ points. He means it literally. The same mind, given a better place to stand, becomes capable of things it could not do a moment earlier. This is why the useful work is so often not new capability but bringing existing capability within reach of an ordinary person.
And the same thing again, in a machine
There is a reason this essay is named after a paper.
A language model without context is not stupid. It is a very confident stranger. It will answer, fluently, having no idea what room it is standing in. Give it the same question with the surrounding facts attached and the answer changes completely — not because the model got better, but because it can finally see what is in front of it.
Standing in that gallery, bored, I was running with an empty context window.
The word does the same work in both places, and it is not a coincidence. A thing on its own means very little. Almost everything a person understands, they understand because of what it is sitting next to.
What this costs
The mechanism that makes a Rothko legible is the same mechanism that makes an empty canvas sellable. That is uncomfortable and it should be said plainly, because in technology most of what is being described is invisible — a protocol, an architecture, a boundary — and there is nothing to check the story against.
So there has to be a test, and it is a hard one:
After the context, can you see it yourself?
If she had walked away and I could still find my way around that room, I was given an explanation. If I could only repeat what she said, I was sold something. A good explanation leaves you more able to look. A bad one leaves you dependent on the person who gave it to you.
That test applies to this website, and to the papers on it, and I would rather you ran it than took my word for anything.
The part I have not solved
She was standing next to me.
She could see the moment I stopped listening. She could change tack, point at a different painting, drop the history and say something simpler. A live explanation with a feedback loop of about two seconds.
Writing has none of that. You set the context down once, for someone you will never meet, whose starting point you do not know, and you never learn whether it worked. Every whitepaper, every protocol spec, every homepage is an attempt to do for a stranger what she did for me in a gallery, without being able to watch their face.
I do not think a good sentence closes that gap. But knowing the gap is there is the difference between writing to explain and writing to impress, and only one of those survives you leaving the room.