AI & Machine Learning Tokenization & Embeddings Quiz

Before a language model can read anything, text is split into tokens and each token is mapped to a vector. This quiz checks how subword tokenizers like BPE work, what embeddings are and why they explain some surprising LLM behaviour.

Medium 10 questions ~6 minutes Pass at 80%

Choose a mode

Playing as a guest: sign in to keep your XP and streak.

Sample question

Question 1 of 10

Why do most modern language models use subword tokenization instead of one token per whole word?

Show the answer

B. A fixed-size vocabulary can still represent rare and unseen words by splitting them into known pieces A whole-word vocabulary would be huge and still miss new words, typos and names. Subwords keep the vocabulary manageable: common words stay one token while rare ones split into pieces like token + ization.

What this quiz covers

  • Why subword tokenization beats whole words and single characters
  • How byte pair encoding (BPE) builds its vocabulary
  • Embeddings as learned dense vectors and cosine similarity
  • Static vs contextual embeddings
  • Token quirks such as letter counting and non-English text

Questions people ask

Do I need an account?

No. Anyone can play. Sign in to keep your XP, streak, badges and history.

Can I retake it?

Yes. Your best score is kept, and the questions and answers are shuffled each time.