Karpathy Resources
This list keeps the Karpathy material in the right places instead of treating every video like required homework.
1. The Spelled-Out Intro To Neural Networks And Backpropagation
- Best time to use: Week 3
- Why it matters: This is the best bridge between “I can run code” and “I actually understand gradients.”
- What to focus on:
- the
Value object
- local gradients
- chain rule
- topological sorting
2. makemore Part 1
- Best time to use: Week 4
- Why it matters: It shows how a tiny language model can still teach huge ideas.
- What to focus on:
- character vocabularies
- counts and probabilities
- sampling
- why generated text feels “kind of right”
3. Let’s Build GPT: From Scratch, In Code, Spelled Out
- Best time to use: Weeks 6 and 7
- Why it matters: It shows the full pipeline mindset: data prep, model shapes, training logic, and generation.
- What to focus on:
- batching
- tokenization
- context windows
- the difference between training and inference
4. Zero To Hero Playlist
- Best time to use: As a reference all course long
- Why it matters: It gives a broader map of what the student is building toward.
- Suggested use:
- do not binge the whole playlist
- watch only the video that matches the current week
- pause and reproduce one small idea immediately
5. Neural Networks: Zero To Hero Notes Strategy
Use these questions while watching:
- What is the main object or data structure in this lesson?
- What gets updated during training?
- What is the smallest thing I can rebuild myself after this?
- What is still fuzzy?
When To Use Non-Karpathy Resources Instead
Karpathy is excellent for build intuition, but sometimes another teacher is better for a single concept:
- 3Blue1Brown: visual intuition for neural nets and gradients
- StatQuest: simple, step-by-step explanations
- PyTorch official tutorials: framework-specific details
- OpenAI or other official docs: API usage and security practices
The goal is understanding, not loyalty to a single teacher.