🏋️ Model Training

5th Grade 🤖 Artificial Intelligence

Model training is the process in which an AI system learns from data. At the end of training you get a model — a program that can make predictions or give answers based on what it learned.

Free · US Letter · no sign-up
Google Classroom Save to Pinterest Email me

At the start of training, the model guesses almost at random. Each time, it gets an example, guesses an answer, and then we measure how wrong it was. Based on the size of the mistake, the numbers inside it are changed a little so it will guess better next time. This is repeated a huge number of times — sometimes millions — until the model gets much better.

Training big models takes lots of data, powerful computers and a long time, which also means a lot of electricity. Once training is done, using the model is much faster. A model trained on one thing — say, recognizing flowers — usually can’t do something else, like recognizing cars, without more training.

A common mistake is thinking a model keeps learning on its own from every chat or use. Most of the time, a model doesn’t change while people use it; it changes only when its developers train it again. That’s why a model doesn’t “know” on its own about things that happened after its data was collected — unless the tool looks up current information, for example on the web, and adds it to the conversation.

For example

  • A model is trained on thousands of pictures of handwritten digits so it can read numbers on forms.
  • A model is trained on speech recordings plus their written text so it can turn speech into writing.
  • A trained model is tested on new pictures to see how often it’s right.
  • When developers find that a model makes lots of mistakes on one kind of example, they add more examples like that and train it again.

📌 Remember: Training = guess, measure the mistake, make a small fix — over and over.

Quick quiz

  1. What happens in each step of training a model?

  2. What does it take to train a big model?

  3. Does a model know about things that happened after its data was collected?

🔎 Ask Thinkapedia another question