It's clickbait. They don't do anything like prove the platonic representation hypothesis.

They do seem to have some really interesting training efficiency improvements, though. The paper is at least worth a read.