verbatome
Verbatome / 2026From text to assistant

Inside
an LLM

Essential language-model mechanisms become objects you can observe, test, and change.

01-08

A progressive path from raw text to the assistant. No AI background required. You don’t need to be a mathematician.

Start the course

Try it now

Break a sentence.

Type to see the units produced by our teaching tokenizer.
35 / 240 characters
A␣model␣receives␣tokens,␣not␣words.

21 tokens

How we learn

Understand it with your hands.

  1. 01

    Manipulate before naming

    Start with a concrete example, risk a prediction, then manipulate the mechanism. The explanation makes sense of what you just saw.

  2. 02

    Build without black boxes

    Each lesson picks up where the previous one stopped: text, tokens, vectors, context, blocks, prediction, training, then assistant.

  3. 03

    Watch the computation take shape

    Tokens, vectors, weights, and probabilities stay visible while you manipulate them.

Index / 08

The learning path

A progressive path from raw text to the assistant. No AI background required. You don’t need to be a mathematician.

01

Text → prediction

  1. The game a language model plays

    No neurons yet: count what follows each word and run the generation loop yourself.

02

Text → tokens

  1. Tokenization, or how a model reads

    Characters, words, or subwords: three ways to break a sentence, three different bills.

03

Tokens → vectors

  1. Embeddings: meaning becomes geometry

    Drag words around a toy map and watch resemblance become an angle.

04

Vectors → context

  1. Attention, where tokens start talking

    Same word, two sentences: watch attention build two different vectors.

05

Context → block

  1. The Transformer block, the floor a model stacks

    Positions, normalization, MLP, residuals: follow one vector through a complete floor.

06

Block → next token

  1. Logits, temperature, and the token that comes out

    The final vector becomes a bet over the whole vocabulary - and the untrained toy has a surprise in store.

07

Error → learning

  1. Training, the machine that fixes its own numbers

    Loss, gradient, small step, repeat: watch chapter 6’s absurd bet flip.

08

Model → assistant

  1. From raw continuer to assistant

    Same tower, three schools: base, exemplary dialogues, human preferences - and their limits.

Step inside the machine.

Open the first chapter