I also found his ‘visual proof’ for neural networks as general function approximators super intuitive.