Lesson 3: What is Inference? — The Intuition, No Code

You don't need to code or know anything about GPUs to start — this lesson has zero lines of code. We begin from the big idea: an AI model has two lives. First you train it (once, slowly), then you run it over and over to get answers. That running is called inference, and it's what this whole course

Training is learning a recipe by tasting and fixing, once. Inference is cooking that finished recipe over and over — without changing it, as fast as possible.

Inference: The stage where you run an already-trained model on new input to get a prediction. Nothing in the model changes — you just get an answer.
Training: The learning stage: the model sees many labeled examples, makes mistakes, and corrects itself until it 'knows'. It happens once and takes time.
Prediction: The answer the model returns for new input — for example 'spam' or 'not spam', 'cat' or 'dog'.