Lab
AI Brain Lab
A chatbot does three simple things over and over. It cuts your text into tokens, it scores which piece could come next, and it picks one. Each station lets you do one of those steps with your own hands.
Kit
I set up three stations for you. Go in order the first time. Nothing you type leaves this browser.
Station 01
Cut text into tokens
A model reads text as tokens: chunks of text, each with an ID number. Type anything below and watch a real tokenizer cut it up.
Tokenizer: o200k_base, the one used by GPT-4o and several newer OpenAI models.
- Tokens
- …
- Characters
- 72
- Words
- 16
- Tokens per word
- …
Tokens of your text
Loading the tokenizer (about 1 MB, only the first time)…
Tokens matter because models are priced and limited by them. Vietnamese often costs more tokens per word than English, since tokenizers learn their chunks from text that is mostly English.
Same meaning, two languages
Each pair says the same thing. Compare how many tokens each side costs.
Greeting
Good morning! Did you sleep well?
English: …
Chào buổi sáng! Bạn ngủ có ngon không?
Vietnamese: …
Coffee
Could I have an iced milk coffee, please?
English: …
Cho mình một ly cà phê sữa đá nhé.
Vietnamese: …
Weather
Bring a raincoat, it's going to rain.
English: …
Nhớ mang áo mưa nhé, trời sắp mưa rồi.
Vietnamese: …
AI
Artificial intelligence learns patterns from data.
English: …
Trí tuệ nhân tạo học các quy luật từ dữ liệu.
Vietnamese: …
Tết
My grandmother wraps sticky rice cakes for Tết.
English: …
Bà mình gói bánh chưng để ăn Tết.
Vietnamese: …