As they say explicitly in the article, this is like criticizing the first rocket to reach the edge of space for not getting anywhere useful.

Yes, this is exactly what it is doing [1]. The area of research is Noisy Intermediate Scale Quantum, and has arisen specifically to prove that quantum supremacy is possible in practice. It is currently the focus of pretty much all quantum computing research because attempts to produce a generalized quantum computer have all failed miserably. Existing practical quantum computers (like DWave) perform various annealing tasks but have basically proven to be inferior to probablistic algorithms computing the same task.

To date all attempt to produce valid claims of quantum supremacy via this channel have failed on closer inspection, and there is no reason to assume otherwise in this case until researchers have had time to look at the paper. There's a number of skeptics in the quantum computing field that believe that this is simply not possible.

[1] https://news.ycombinator.com/item?id=42369463

It's different from your hourglass in that the computer is controllable. Each sampled random circuit requires choosing all of the operations that the computer will perform. You have no control over what operation the hourglass does.

It won't be factoring large numbers yet because that computation requires the ability to perform millions of operations on thousands of qubits without any errors. You need very good error correction to do that, but luckily that's the other thing they demonstrated. Only when they do error correction, they are basically combining their system down into one effective qubit. They'll need to scale by several orders of magnitude to have hundreds of error corrected qubits to do factoring.