How does an LLM chatbot work?

From the user's message to the generated response — an animated walkthrough of each step inside a large language model.

Step 1

User Input

The user types a message. The model receives raw text.

Step 2

Tokenisation

Text is split into sub-word tokens, each mapped to an integer ID.

Step 3

Embeddings

Each token ID becomes a dense vector + positional encoding.

Step 4

Self-Attention

Each token attends to all others. Repeated across N transformer layers.

Step 5

Generation

Output probabilities → next token selected → word-by-word generation.