I like all the amazing comments and critique there but have we considered that somehow, the neural network was able to formulate the concept of gravity, emulate a chaotic system and do all of it using a constant computational cost? It is a bunch of interacting units that reconfigured itself using gradient descent to behave like a chaotic three body system. I find that quite fascinating. Although I would agree that the hype created is not justified but this neural network approach bypasses concepts of gravity, newtonian mechanics, chaos, ODE integration and gives a simple function to calculate position at arbitrary times. Quite interesting.