By Team Edtoolkits · AI learning
Compare normalized next-word probabilities with the choices sampled from them.
Use the guide alongside the working tool.
Open Next-Token Prediction →Explore the lesson
Watch the animation pick the most likely next word, again and again.
In the playground, click a candidate to add it, or let the model pick with the temperature you set.
Flip the memory card and answer the quick check.
A comparison to try
Choose a starting word, inspect the candidates and continue repeatedly. Compare lower and higher temperature with that same starting context.
Read the result
Temperature reshapes the distribution. Choosing a less frequent candidate does not contradict that distribution; it is part of sampling from it.
A mistake to avoid
This tiny model remembers word pairs rather than understanding a whole sentence. A likely continuation need not be true, helpful or responsive to a question.
Teaching model and privacy
The playground uses a tiny word-pair model built from a few dozen sentences, so it shows the idea, not the quality of a real LLM. Everything runs in your browser.
Ready to apply this workflow?
Try Next-Token Prediction →Explore a related workflow
- Choose a course entry point and verify saved progress
Use prior knowledge to select a starting chapter while distinguishing local progress from an account or a qualification.
- AI Tokens: Inspect Pieces, Spaces and Stable Teaching IDs
Compare common words, rare words and numbers without using the playground as an exact provider counter.
- Choose complexity using held-out error
Compare training and test curves before calling a detailed fit an improvement.
- Bias and Variance in your browser
Throw model 'darts' on a target: stiff models land together but off-centre, wiggly ones scatter. Slide the complexity and see bias and variance.
Found unclear instructions or an unexpected result? Contact Team Edtoolkits.