Intelligence is being built faster than it is understood.
The dominant method works. Add data, add compute, optimize until the answers look right. It has produced systems of extraordinary capability, and it will produce more.
It has not produced understanding. What a trained model represents, the computations those representations support, and why a capability appears when it does are largely unknown. We measure behavior because we cannot yet read mechanism.
A trained model is not a finished product.
It is an unexamined object.
Consider how a physicist works through a problem in mechanics. They do not search a list of equations. They can see what the system is doing, and the right law is obvious before it is written down. That is what intuition actually is: not a guess, but the felt shape of principles that are genuinely understood.
We want that footing with learned systems. Superintuition studies what training creates, from the inside: the structures a network forms, the algorithms it implements, and the mechanisms that produce generalization. We are after explanations that are causal, predictive, and precise enough to act on.
This matters for reliability, control, and safety. It matters just as much for capability. A field that understands its own artifacts can design them, instead of discovering them by accident.
We begin with artificial intelligence. We do not assume the questions end there. Intelligence is not only a property of machines, and the principles that explain one system may explain others.
We are a small team of researchers and engineers working to understand intelligence from first principles.
We intend to understand it well enough to build it deliberately.