I thought the same thing. Surely a random binary sequence will have some (small) repeating sequences? So as long as we can find them and (efficiently) mark their locations, we can have some small size reduction.

Which is basically what he’s done here. He’s cheating by marking the location of the repeating sequence using unique files, rather than some other actually more efficient location system.

This is a fun discussion! I think it helps to make compression feel more approachable for armchair enthusiasts.