Typical clickbait title. The puzzle hasn’t been “cracked”, someone who knows the problem domain looked at how neural nets trying to approximate the puzzle, said “why don’t we try it this way”, and suddenly their approximations got way better.
ArXiv Link for folks interested in the actual paper.
They claim PDEs are notoriously hard to solve, then latch onto one of the hardest. The Navier-Stokes equation is hard because it’s nonlinear, not because it’s a PDE. Linear PDEs, such as variations of Laplacians or wave equations, are often straightforward to solve numerically and even analytically in many cases.