I was debugging something last week and had this weird moment where I could tell immediately that the output was wrong but had no idea how to fix it. Sat there for like twenty minutes just staring at the error, knowing exactly what wasn’t working and zero clue what would.
And I realized — this isn’t just a programming thing.
It shows up everywhere once you look. In science: disproving a hypothesis takes one counterexample, proving it takes forever. Learning a language: you can hear bad grammar way before you can produce good grammar. I spent a year in France and by month three I could tell when someone sounded off. By month twelve I still couldn’t order food without getting laughed at.
The gap is checking versus building. Verifying is cheap. Constructing is expensive. And yet somehow we cross this gap all the time — we use wrongness as a compass, slowly walking from “that’s not it” toward “this might be it.”
I’ve been trying to find what people call this. Optimization? Solomonoff induction? Generators vs discriminators? None of them quite fit what I’m pointing at. Feels like there should be a name for the specific thing where verification is the cheap half of learning and construction is the expensive one.
Or maybe I’m just describing something obvious and don’t know the word for it. That happens a lot.
I was debugging something last week and had this weird moment where I could tell immediately that the output was wrong but had no idea how to fix it. Sat there for like twenty minutes just staring at the error, knowing exactly what wasn’t working and zero clue what would.
And I realized — this isn’t just a programming thing.
It shows up everywhere once you look. In science: disproving a hypothesis takes one counterexample, proving it takes forever. Learning a language: you can hear bad grammar way before you can produce good grammar. I spent a year in France and by month three I could tell when someone sounded off. By month twelve I still couldn’t order food without getting laughed at.
The gap is checking versus building. Verifying is cheap. Constructing is expensive. And yet somehow we cross this gap all the time — we use wrongness as a compass, slowly walking from “that’s not it” toward “this might be it.”
I’ve been trying to find what people call this. Optimization? Solomonoff induction? Generators vs discriminators? None of them quite fit what I’m pointing at. Feels like there should be a name for the specific thing where verification is the cheap half of learning and construction is the expensive one.
Or maybe I’m just describing something obvious and don’t know the word for it. That happens a lot.