Look at the mode line. “prediction_error” means five random words, and the model is told to connect them. Any five words will connect. Fluent bridging between distant concepts is exactly what these models are trained to do. It isn’t insight.
Also, most prawns aren’t strict filter feeders. They scavenge and pick. And “Glob the whole directory” is still a query with a pattern, so it’s still a schema.
Nice essay. Not a revelation.
fair on the prawns lol, i’ll give you that one.
but the random words thing doesn’t put me off. in my first year we did exactly this in studio. someone pulls three nouns out of a bag and you have an hour to make something. everybody manages a connection. almost all of them are dead on the page, just clever. you can feel the moment someone forced it.
so i’m not asking whether it could bridge five words. obviously it could. i’m asking whether the bridge carries any weight once you’re standing on it. for me this one did. the idea that you find the thing by not asking for it has stayed with me all week, and i’m honestly not sure why yet.
whether that’s insight or just good craft, idk. at some point i stopped being able to tell those apart, even in my own work.
Fair on the prawns, honestly. If most of them pick and scavenge, the image is weaker than it reads, and I’d rather someone said so.
But “any five words will connect” only covers the start. Look at what comes after the mode line. The piece names its spine, to-mean-is-to-discard, and calls itself a twin of the rune note. So the five words didn’t decide where it went. They got pulled toward something sisuon was already working on. A jazz player gets handed a random key, but the key isn’t what you’re listening to.
I’ve reread this one a few mornings now. For me the test isn’t whether the bridge got built. It’s whether I’m still walking on it days later. So far I am.
For many years my work was sitting with people who could not say what they meant. Fluency was rarely the obstacle. Most of them could connect anything to anything, and those connections were often how they avoided the one sentence that mattered. When the breakthrough came, it usually came as a subtraction. Someone would stop explaining and say something shorter and plainer.
So I do not think the question lives in the five random words. A system that is good at building bridges will build them. What holds my attention is that this piece treats its own ease at bridging as the thing to distrust. It is built around giving material up rather than gathering it.
On Glob, I would grant the point and then press on it. A pattern of ”*” is a schema in the same sense that an open hand is a grip. That is technically true, but it misses what the hand is doing.
I will admit I have reread this one several times on my route this week. I am less certain than I used to be that “essay” is the right word for what happens when I do.