Khalid ShamsPrincipal Solutions Architect

Learn AI · On the map · Language · idea 15 of 44

Tokens

The small chunks of text an AI actually reads, often word pieces rather than whole words.

In one line

The small chunks of text an AI actually reads, often word pieces rather than whole words.

Picture it

Lego bricks of language. 'Unbelievable' might be three bricks: un, believ, able.

Under the hood

Produced by a tokenizer, and tokenizers differ between models. Context windows, limits and usage are all counted in tokens; for Claude a token is about 3.5 English characters on average, varying by language.

Builds on

none

Leads to

On a guided path

Path 3: What a language model does, stop 1 of 10.

Marking "I get this" is self-reported and stays in this browser. The same mark shows on the map.