The paper claims that a classical supercomputer is MUCH slower at simulating a noiseless 53 logical qubit system, but doesn't exclude the possibility of a class

The paper claims that a classical supercomputer is MUCH slower at simulating a noiseless 53 logical qubit system, but doesn't exclude the possibility of a classical supercomputer simulating an approximation of the actual noisy 53 bit physical qubit system, that still gives sufficiently decent estimates of the observed final state probabilities. Such a simulation would compute orders of magnitude fewer than 2^53 amplitudes, and use whatever shortcuts are available to get approximately correct results. Quantum supremacy cannot be claimed until this possibility is ruled out.