Miatz playground

Token Shredder

Type a word, watch AI shred it into tokens.

AIE-101Browser (JS)AI-Concept-Vizdev · data
The Demystify signature

The 'why AI can't spell' reveal: proves, on the learner's own typed word, that the model operates on opaque token IDs, not letters -- the mechanism behind a famous, real failure mode, not a claim about it.

How it works

What goes in, what comes out

What it does

Learner types any text and it splits live into color-coded token chips, each labeled with its numeric ID as the model itself would see it. A side-by-side 'letter view' shows the same text as individual characters, so the gap between what a human reads and what the model actually receives is visible at a glance, not asserted.

You bring

Free text typed or pasted by the learner (or one of three preset 'gotcha' strings, incl. 'strawberry').

You get

A color-chunked token strip, live token count, per-1K-token cost estimate, and a one-line reveal of why the model can't count letters in a word it never sees as letters.

You control

Tokenizer/model-family selector; language toggle (English vs Hindi/Tamil/Spanish to expose non-English token bloat); show/hide token IDs; running token-count and estimated-cost readout.

Where it's used

Free lead magnet at /playgrounds and embedded on resource/blog pages (GEO/SEO); paired demo inside AIE-101's opening tokens-and-vocabulary lesson.

tokenizationsubword vocabularytoken IDsnon-English token cost

Concepts click when you open the machinery.

Three labs are already live and free — the same hands-on style this playground brings to its module.