Everyone is getting better at using AI. I want to understand how it's made.

Every major company became a software company at some point, or died. The same thing is happening with AI, and knowing how to prompt isn't the same as understanding how these systems work.

Prototypes and vibe-coded projects can stop at the API layer. Products that need to compete can't. To build at that level, you need to understand what's below the API. So I asked Claude to put together a 54-week syllabus for me, scoped as if I was becoming an AI research engineer.

Here's what it covers:

• Months 1-3: Math foundations, core ML theory, deep learning fundamentals. Build a mini-GPT from scratch.

• Months 4-5: Survey sprints across reinforcement learning, language model training pipelines, and diffusion models.

• Months 6-8: Pick a specialization and replicate a research paper end-to-end.

• Months 9-11: Open-source contributions, a second research project, and technical writing.

• Months 12-13: Frontier topics (interpretability, scaling laws, AI safety) and a capstone project.

I'm starting this week. Join me.

The full syllabus is on GitHub: https://github.com/michaelcjoseph/ai-syllabus