Step 1
Study deeply
Work through the Deep Learning Specialization rigorously — CNNs, sequence models, optimization.
Advanced · Learning path
Research AI systems
Understand, replicate and advance the state of the art.
At a glance
- Who it’s for: Engineers and researchers who want to read papers fluently, reimplement methods, and contribute at the frontier.
- You’ll be able to: You will be able to read and reproduce current research, build core architectures from scratch, and follow the field at the level where new ideas are formed.
- Time: ~120 hrs · Steps: 5 · Math: Heavy
The deep end. This path assumes comfort with code and a willingness to do the math. You’ll work through rigorous deep-learning theory, a serious NLP course, and the rite of passage of building a GPT from scratch — then make reading the daily arXiv flow a habit so the frontier stops feeling like a moving target.
The path
5 steps · all resources free to start
Step 2
Take a frontier course
Stanford CS224N (NLP) or CS231N (vision) — full lectures and assignments are public.
Step 3
Build GPT from scratch
Karpathy’s Zero to Hero series demystifies backprop through to a working language model.
Step 4
Read papers daily
Make the arXiv flow a habit — our Research feed surfaces and explains the day’s papers.
Step 5
Follow the discourse
The Gradient and Distill.pub keep you close to how the field thinks, not just what it ships.
Get the briefing
The one story that matters, 5 headlines and the paper everyone's citing — every Tuesday, free.
Subscribe free