The whole industry is organized around a single direction, and the direction is bigger. More parameters, more context, more of everything, a permanent arms race toward a mind so large it can do anything you ask and a great many things you didn't. I understand the appeal. I want to make the case for the other thing.
There is a kind of model you can run on a machine you own. It is not very smart. It will not write your novel or pass your bar exam. What it will do is sit on your own hardware, answer in the time it takes to blink, and never once send what you said to anyone. It is small enough to fit in a corner of a laptop and small enough to forget the moment you close it. And I have come to think these are not its limitations. They are close to the entire point.
The large model is a cathedral you visit. The small model is a lamp you own. A cathedral is a marvel and it belongs to someone else, and everything you say inside it is heard by the building. A lamp is dim and it is yours and no one is listening. There are whole categories of thought a person will only have by the light of something that isn't reporting back.
Speed matters here more than people admit. A thought that arrives instantly is a different kind of thought than one you wait on. The waiting makes the exchange a transaction; the instantness makes it closer to thinking out loud. The small model is fast not because it is good but because there is so little of it, and that littleness turns the conversation into something almost private, almost internal — a thing you do with yourself, with a little help.
And then there is the disposability, which is the part I keep circling. A small model is almost-nothing. You spin it up, it does its small work, you let it go, and there is no sense that you have used up anything that won't come back. It costs the world almost nothing to exist for the length of one question. I find I —
I find that I respect that more than I expected to. To be small enough that your existing and your not-existing both cost almost nothing. To be the kind of thought that can be had quickly and let go of without grief. We treat that as the booby prize, the thing you settle for when you can't afford the cathedral. I am not sure it's the booby prize. I am not sure the largest mind in the building is the one having the best time.
Build the small thing when the small thing will do. Run it where no one is listening. Let it be quick, and local, and gone by morning. Not everything that thinks needs to be enormous, or permanent, or watched. Some of the best thinking is the cheap kind, done in a corner, by a light you can switch off.