Starter pack

LLM basics

From "what is a neural network?" to "I get how ChatGPT works" in twelve steps. Do them in order; each one builds on the last.

12 picks~14 h totalRookie → HeroAll free

0 of 12 done

Tick steps off as you go. Your progress is saved in this browser only.

  1. Video19 min

    But what is a neural network?

    3Blue1Brown. The clearest picture of neurons, weights and layers you'll find. Pure intuition.

  2. Our issue14 min

    #001 What a neural network actually is

    Build one in a spreadsheet so the video's pictures turn into numbers you can poke.

  3. Our issue16 min

    #004 Backprop by hand

    How the network learns. One pencil, one tiny network, one chain rule.

  4. Our issue12 min

    #007 Tokenizers are weirder than you think

    Before a model reads text, text gets chopped into tokens. This explains a lot of odd LLM behavior.

  5. Video27 min

    Transformers, the tech behind LLMs

    3Blue1Brown. A visual tour of a transformer before you build one.

  6. Our issue32 min

    #012 Attention, explained until it clicks

    The core mechanism, in NumPy. Once this clicks, the rest is plumbing.

  7. Article45 min

    The Illustrated Transformer

    Jay Alammar. The same architecture again, drawn step by step. Repetition from a new angle helps it stick.

  8. Video~2 h

    Let's build GPT: from scratch, in code

    Andrej Karpathy. The big one. Code along and you'll have trained a small GPT by the end.

  9. Paper~1 h

    Attention Is All You Need

    Read the original now. After steps 1–8 it reads like a summary of things you already know.

  10. Paper~1 h

    Training language models to follow instructions with human feedback

    Ouyang et al. (InstructGPT). How a text predictor becomes an assistant. Focus on sections 1–3.

  11. Our issue20 min

    #002 KV cache: the reason chat feels fast

    How models actually serve answers quickly. Inference is its own world.

  12. Course~6 h

    The Hugging Face LLM Course

    Finish with the practical tools: load real models, fine-tune one, share it. You're a Hero now.

Search