Self-taught AI Master's curriculum published by demirbey05 — a free, 2-year self-learning path from math fundamentals to research-level AI

demirbey05, driven by 6 years of self-learning philosophy, published a structured 2-year AI Master's Curriculum on Hacker News. It requires prerequisites in calculus, linear algebra, algorithms, and programming, then chains Stanford courses (CS231n for computer vision, CS224n for natural language processing, CS229 for machine learning theory) and MIT 6.7960 (deep learning) through Year 1, followed by hands-on specialization tracks in generative AI, healthcare AI, and other domains in Year 2.

The curriculum treats traditional university education as outdated for AI—arguing that knowledge transfer should come from great teachers, not institution prestige. Year 1 covers foundational theory (how neural networks and optimization actually work), while Year 2 flips to applied research (building diffusion models, fine-tuning large language models, designing medical imaging systems), mirroring how industry AI teams are structured.

For anyone learning AI outside university—bootcamp graduates, career-switchers, software engineers—this gives a concrete 24-month roadmap with named courses and a rationale for each stage, replacing the common trap of jumping between YouTube tutorials and Medium articles. Students and self-taught practitioners can now benchmark their learning against a coherent, SOTA-aligned plan rather than guessing what to study next.