Maia

Ready to start

LLMs

Large Language Models

Tokens, pretraining, prompting, tools, evaluation. Later headings open after the one before. Explain tokens, pretraining, prompting, finetuning, RAG, and evaluation well enough to design and debug an LLM system.

  • 6 headings
  • Ladder
  • Test out a heading you have
Start Large Language Models free

Free to start. Sign in with Google or email.

A class on this path

Tokens and prediction · First class

Live tutor
Materials for this classReadingPodcastLecture

Pick where you want to start

Each heading is its own course. Later ones open when you finish — or skip — the one before. A later course does not reteach earlier bands.

  1. Tokens and prediction

    Start here

    Tokenise a sentence, and explain next-token prediction and sampling vs greedy with a tiny example.

  2. Pretraining

    After Tokens and prediction — test out to unlock

    Say what the pretraining objective is, what data does, and what a base model still cannot do.

  3. Prompting

    After Pretraining — test out to unlock

    Write a prompt with spec, examples, and checks, and know when prompting is the wrong tool.

  4. Finetuning and preference

    After Prompting — test out to unlock

    Separate SFT from preference / RLHF-style training, and say what each changes in behaviour.

  5. RAG and tools

    After Finetuning and preference — test out to unlock

    Design a retrieve-then-read or tool loop, and name the failure (wrong doc, stale, hallucinated cite).

  6. Evaluation and failure

    After RAG and tools — test out to unlock

    Pick a metric or eval set that would catch the bug you care about, including jailbreak and leakage.

What this is for

  1. 01

    Tokenise a sentence, and explain next-token prediction and sampling vs greedy with a tiny example.

  2. 02

    Say what the pretraining objective is, what data does, and what a base model still cannot do.

  3. 03

    Write a prompt with spec, examples, and checks, and know when prompting is the wrong tool.

Why a path, not a playlist

Tokens, pretraining, prompting, tools, evaluation. Later headings open after the one before.

Each heading is its own course with a tutor. Later headings open when you finish — or test out of — the one before.

After you sign in

  • You land on this map — every heading is already there.
  • Take the open class, or test out of a later heading to unlock it.
  • Each heading is one shared course — same syllabus and first class for everyone. Later classes continue on your copy.

LLMs

Start Large Language Models free

Free to start. Sign in with Google or email.