In the machine learning, there are a few closely guarded secrets that experts rarely discuss openly. One of the biggest is that we don't fully understand why deep neural networks (DNNs) work so well. The lack of theoretical consensus often forces us to treat AI as a black box, resorting to trial-and-error approaches when fine-tuning parameters. In this presentation, Birdman will take a casual yet insightful approach to guide the audience through the development of machine learning, current theoretical frameworks, and the dynamic systems of neural networks. He will also share his personal learning experiences and unique perspectives, hoping to inspire more people to explore the underlying truths behind machine learning algorithms. (This talk is a refined version from CraftCON 2024.)