Most efforts to build scientific AI aim to make science faster: more experiments run, more literature searched, more of the hypothesis space explored. This is worth doing, but it mistakes the prize. The discoveries that reshape a field were never faster moves inside an existing framework; they were the moments the framework itself was redrawn. Copernicus did not compute better epicycles. Einstein did not refine the ether. Progress at the frontier is not interpolation. It is the replacement of one way of seeing the world with another.

We build machines for that kind of progress: not systems that do more normal science, but systems that can do revolutionary science, generating new paradigms rather than recombining old ones.

Scientific superintelligence, to us, is not an oracle that answers faster. It is a system that changes what the questions are.

Get in touch →