Andrej Karpathy
Neural Networks: Zero to Hero
Builds backpropagation, then a language model, then a GPT, from scratch in a notebook. Nothing is imported that has not been derived.
Who it suits
The clearest explanation of what a transformer actually does. Not a route to employment on its own.
What it covers
Deep learningLanguage models
What is not recorded here
No rating. Course ratings change weekly, differ between platforms for the same material, and are dominated by people who finished, which selects for those the course already suited. Level, duration and cost are stable and checkable; the line above is the judgement a rating stands in for.
Concepts this covers
TransformerTokenEmbeddingsAttentionLarge Language Model (LLM)Retrieval-Augmented Generation (RAG)
Related courses
- Practical Deep Learning for Coders · fast.ai, ~7 weeks
- Introduction to Deep Learning, 6.S191 · MIT, ~10 lectures
- Hugging Face LLM Course · Hugging Face, 12 chapters
- Building with the Claude API · Anthropic Academy, 84 lessons, 8+ hours
Common questions
Is Neural Networks: Zero to Hero free?
Yes, entirely. Provider: Andrej Karpathy.
How long does Neural Networks: Zero to Hero take?
Roughly ~15 hours, at the pace the material is designed for. Anyone working full time should expect longer.
What level is Neural Networks: Zero to Hero?
Intermediate. Builds backpropagation, then a language model, then a GPT, from scratch in a notebook. Nothing is imported that has not been derived.
Who is Neural Networks: Zero to Hero for?
The clearest explanation of what a transformer actually does. Not a route to employment on its own.