Predicting the next token

Year 7 · Lesson 4 · Content 2026.11.02.1

All supplied-record lessons · Open interactive lesson

Supplied-record alternative. Enter your responses in your school assignment. These supplied results support learning but do not show that you ran an algorithm.

Your goal

Explain why a likely next word does not prove a fact.

To be able to explain how next token prediction can produce fluent text that still needs fact checking.

Before you start

Complete “Please close the __”. Choose one possible continuation.

Choices: door / window / book

Bridge feedback after trying

Several continuations make sense. A likely ending is a language prediction; it does not tell us which event happened.

Words for this lesson

Token
A small text unit such as a word or part of a word.
Probability
How likely a possible choice is.
Fluent
Written or spoken so the language flows smoothly.

The idea

A language model uses earlier text to predict possible next tokens. It selects one and repeats the process. A token can be a word or part of a word. The most likely choice is not the only possible choice. This process can make fluent text, but an important factual claim still needs evidence. Some AI services add search or other tools; check their actual sources too.

A completed example

Practice: a toy animal model gives cat seven of ten possible draw numbers and dog three. Draw 9 selects dog. Cat was more likely, but dog was still possible. This teaches selection, not whether a real animal was present.

Check your understanding

A token can be ...

  1. A word or word part
  2. Only a full paragraph
  3. Only a true fact
Practice feedback after trying

A word or word part. Tokens are small text units, not truth labels.

50% next-word probability means ...

  1. A likely next choice in this model
  2. The whole sentence is 50% true
  3. A verified source
Practice feedback after trying

A likely next choice in this model. Choice probability does not measure factual correctness.

A missing award entry proves ...

  1. These records do not establish the award
  2. The award definitely happened
  3. The award could never exist
Practice feedback after trying

These records do not establish the award. Say what the evidence permits.

Task 1 · Make a prediction

Every draw must give green

Options: Yes / No / Not sure

Write your prediction, or state that you skipped it.

Try it using supplied records

  1. Read the three colour choices and their chances. Predict whether every draw must give green.
  2. Press Draw next token four times. Use the fixed draws 2, 7, 10 and 5.
  3. Record the four selected words. Explain why blue and red remain possible.
  4. Open the fictional school records and inspect the first claim with its source.

Scripted probability simulation · not a live language model

Complete data and source records

Word selection. Fictional teaching data.
Draw rangeTokenChance
1 to 5green50%
6 to 8blue30%
9 to 10red20%
Read as stacked records
Draw range
1 to 5
Token
green
Chance
50%
Draw range
6 to 8
Token
blue
Chance
30%
Draw range
9 to 10
Token
red
Chance
20%
Fictional records. Fictional teaching data.
RecordContent
ADune School library opened in 2021. It has 248 science books. It opens at 08:00 on school days.
BA reading club meets on Tuesdays. This events record gives no award information.
Read as stacked records
Record
A
Content
Dune School library opened in 2021. It has 248 science books. It opens at 08:00 on school days.
Record
B
Content
A reading club meets on Tuesdays. This events record gives no award information.
Claims to check. Fictional teaching data.
IDClaim
C1The library opened in 2020.
C2It has 248 science books.
C3It opens at 08:00 on school days.
C4It won the Gulf Reading Star Award in 2024.
Read as stacked records
ID
C1
Claim
The library opened in 2020.
ID
C2
Claim
It has 248 science books.
ID
C3
Claim
It opens at 08:00 on school days.
ID
C4
Claim
It won the Gulf Reading Star Award in 2024.

Expected results · supplied

These are supplied reference results. They are separate from your own explanation. No code needs to run to read them.

1. Fixed draw queue [2,7,10,5] gives green,blue,red,green. Reset restores the queue. No random number generation is needed in the assessed path. One whole colour word is one token only in this toy example.

2. C1 Contradicted: A gives 2021. C2 and C3 Supported by A. C4 Not established: neither record verifies it. Missing information is not automatically proof that an event never happened.

3. This is a scripted probability simulation and an authored generated-style paragraph, not a live language model. Display that label before the interaction.

Help with this step

Hint 1

Find the draw range that contains the number.

Hint 2

A choice with fewer numbers is less likely, but it can still be selected. For claims, compare the exact fact with the source.

Hint 3

Practice: if a timetable says Thursday and a reply says Friday, the reply is contradicted. If the timetable lists no day, the claim is not established.

Task 2 · Use the evidence

  1. Judge the four claims as Supported, Contradicted or Not established from these records.
  2. Write two sentences using supported facts and correct the wrong year.
  3. Explain why smooth wording could not verify the award claim.

My explanation

Write in your school assignment. Keep these answers in your own words.

C1

C2

C3

C4

Your explanation

Explain a fresh example

A fluent answer gives a precise library opening year. How might the text have been produced, and how would you check the year?

My confidence

I need another explanation / I can explain with help / I can explain with my own example

My next step

Write what you will do next.

Optional depth

Why is asking the same model to repeat the year weaker than checking the register? Does this mean every generated sentence is false?

What this activity can show

The colour probabilities and school records are invented. Text token prediction is not a complete explanation of all image and sound generation.

This is a student learning record. Submission is completed through the school assignment system.