Interestingly, all envogue tech of today like ML and Bitcoin converge around this topic. In essence, PoW is just (partly) figuring out the seed of a RNG, which in return is compression at its finest.

Thus, the author has a point in saying "Basically, once we’re done w/ statistics, we’re done. But this mentality may be one of our biggest faults". Even seemingly random data can be represented by incredibly short descriptions (as in MDL). I'm fascinated by compression since I was a child hacking in x86 assembler.