The post discusses the journey of a teenager who, after becoming unemployed, sought to deepen their understanding of deep learning from a systems perspective by exploring deep learning compilers like Tinygrad and ultimately discovering a Common Lisp-based compiler that can run GPT-2. Despite initial challenges with code readability, the project shows promise for future development, including potential support for CUDA. The author views this endeavor as both a fascinating learning tool and a step towards expanding the capabilities of deep learning compilers.