Skip to content
Codextroop TechnologiesCodexLab

How Models Learn

Training shows the model billions of examples, adjusts its internal settings when guesses are wrong, and tests on new text.

Class 9 to 10beginner18 min3 lessons
By the end you will be able to
  • Describe training as examples plus feedback on wrong guesses
  • Explain train and test split in plain words
  • State that training data shapes model behaviour

Finish first

Guessing the Next Word
Topic contents

Lesson 1 of 3

Learning from examples

Before a chatbot talks to you, it trains on huge text collections: books, websites, forums and more. The task is often predict the next token in each snippet.

When the guess is wrong, training nudges internal settings slightly so similar situations score better next time. Many nudges over many examples build skill.

Follow the step-flow animation. See examples enter, guesses happen, feedback arrives, settings adjust, repeat.

Continue

Lesson 2 of 3

Train versus test

Builders hold back some text for testing only. If the model memorised training lines, it would ace training but fail on fresh paragraphs.

Good test scores suggest the model learned general patterns. Bad test scores suggest memorising or missing skills.

You never see the raw training set in class. You interact with the finished model whose behaviour was shaped by it.

Continue

Lesson 3 of 3

Data leaves fingerprints

Models reflect what was common in training. If certain views dominated, outputs may echo them. Rare languages or names may get weak treatment.

Fine-tuning later can adjust a model for a calmer school tutoring tone or stricter safety rules, but it still builds on the base training.

Next you will study confident wrong answers, bias and how to use AI carefully.

Continue

Practice

Work these out yourself

No answer key here on purpose: these are the questions worth thinking through before you move on. Open one and work it out.

0/3 attempted

Assessment

Check your understanding

Answer each question, then read the explanation. That is where the learning is.

0/4
  1. Question 1: Training mostly teaches models to do what?
    Question 1 / 4

    Training mostly teaches models to do what?

    Select an option first
  2. Question 2: Why keep a test set separate?
    Question 2 / 4

    Why keep a test set separate?

    Select an option first
  3. Question 3: When guesses are wrong during training, what happens?
    Question 3 / 4

    When guesses are wrong during training, what happens?

    Select an option first
  4. Question 4: Model behaviour often reflects what?
    Question 4 / 4

    Model behaviour often reflects what?

    Select an option first
4 of 4 questions left.

Finished this topic?

0 of 3 lessons marked done. Marking the topic complete ticks the rest.