Skip to content
Courses

Machine learning

Machine learning has a program learn rules from examples that come with answers, instead of someone writing them, so it can predict new cases.

Select any words here to ask your Maestro about them.

At full detail, the model is 100% right on the members it learned from. How often do you think it's right on members it has never seen?

A decision tree trained on 1,600 gym members and tested on another 400. Example data.

What it's for: predicting something when there are lots of examples with answers but nobody can write down the rules: which customer will cancel, which email is spam, how much will sell. Language models like an LLM are machine learning on text.

How to tell if it works: measure it on data you set aside before training, as in the figure. And what it predicts is a probability, not a certainty: it will be wrong on some cases, which is why you check its output.

Finding that two things move together (correlation) isn't the same as finding a cause: a model learns the first.

Practice checking what a model gives you in When can you trust a number from an AI?.

How it's used

How a model learns

Which gym members will cancel?

  1. Gather examples with answers: 2,000 members from last year and whether they canceled.Columns like visits per month, months as a member and plan.
  2. marked: Set 400 aside for testing and keep 1,600 for learning.The model never sees those 400 while it learns.
  3. The model finds rules: "under 3 visits a month and under 6 months in, likely to cancel".Nobody wrote them: they come from the examples.
  4. Measure on the 400 you set aside: that's the accuracy to expect with new members.
An LLM, like the one behind ChatGPT or Claude, is machine learning on text: it learned to continue text from an enormous number of examples.

When to look closer

When the model is too good

Columns: visits, tenure, plan, cancel_reason

  1. marked: 99% right on the 400 test members

Your model is 99% right on data it never saw. Would you use it tomorrow?

Ask your Maestro about Machine learning

It answers from the free course page closest to this word, the way it answers while you study.

Example

You: What does "Machine learning" mean, in plain words?

Your Maestro: I'll explain it with a case of yours: it sticks better than a definition. What would you predict at work if you had lots of examples?

Asking about: Answers from the page "Someone said "it's in SharePoint": how do I find it?"

Only this browser keeps what you ask here. How it works

Learn more

Related terms

  • LLM (large language model)

    An LLM is a program trained on huge amounts of text to continue a piece of text one chunk at a time: it works out which chunk is likeliest to come next and writes it. ChatGPT, Claude and Gemini all run on an LLM.

  • Probability

    Probability measures how likely something is, from 0 (never) to 1 (certain). Favorable outcomes ÷ possible outcomes: rolling a 6 on a die is 1/6.

  • Correlation

    Correlation measures how closely two numbers move together, from −1 to 1. Near 1: when one rises, so does the other. Near 0: no straight-line link.

  • AI hallucination

    A hallucination is an answer from an AI model that sounds confident and well written but is false or comes from no source at all: a number, a fact or a quote nobody ever said.

The whole glossary