Skip to content
bits
Sign inStart free

Tokens & temperature

Why the same prompt gives different answers — and how to dial randomness up or down on purpose. You'll tokenize sentences, read probability tables and tune a chatbot from boring to unhinged.

  • Beginner
  • 8 lessons
  • ~53 min
  • text + video
Start lesson 1 →Watch the 90-second intro

By the end you'll be able to

  1. Predict roughly how many tokens a prompt will cost.
  2. Explain why a model “chooses” one word over another.
  3. Pick a temperature for a task and defend it.

Syllabus

M1: How text becomes numbers

  1. 1. What is a token? — not startedchatvideo6 min · +30 neurons
  2. 2. Tokenizers are weird — not startedchat5 min · +30 neurons
  3. 3. Counting tokens (and cost) — not startedchat6 min · +40 neurons

M2: Picking the next word

  1. 4. Probabilities, not answers — not startedchat7 min · +40 neurons
  2. 5. Greedy vs. sampled — not startedchat6 min · +40 neurons
  3. 6. Temperature & randomness — not startedchattypevideo7 min · +40 neurons

M3: Putting it together

  1. 7. Top-p and friends — not startedchat6 min · +40 neurons
  2. 8. Checkpoint: tune a chatbot — not startedchatproject10 min · +60 neurons