“Noisy” because even the smallest changes in the environment can lead a quantum system to essentially become random (or “decohere”), and “intermediate scale” because the current generation of quantum computers is still limited to just over a thousand qubits at best. A qubit is the fundamental unit of computing in quantum systems, analogous to a bit in a classic computer, but each qubit can be in multiple states at the same time and doesn’t fall into a specific position until measured, which underlies the potential of quantum to deliver a huge leap in computing power.
It doesn’t matter how many qubits you have, though, if you barely have time to run a basic algorithm before the system becomes too noisy to get a useful result — or any result at all. Combining several different techniques, the team was able to run thousands of experiments with virtually no errors. That involved quite a bit of preparation and pre-selecting systems that already looked to be in good shape for a successful run, but still, that’s a massive improvement from where the industry was just a short while ago. Further reading: Microsoft blog.