The basics
Training is the process of teaching an AI model by showing it many examples and adjusting its internal numbers until its answers become accurate.
Training is where the learning actually happens. The model makes a guess, a loss function measures how wrong the guess was, and a learning rule nudges the model's weights to reduce that error. Then it repeats with the next example. Over millions of rounds, the guesses improve from random to reliable.
Training can be slow and expensive. A small model may train in minutes on a laptop. A large language model can take months on thousands of specialized chips and cost tens of millions of dollars. That is why big models are trained once and then reused by many people.
Training and using a model are separate stages. Once training stops, the model's numbers are frozen. When you chat with it, it is not learning from you in real time. Anything it seems to 'remember' during a conversation is just text kept in its context window.
To build a cat-versus-dog detector, you feed the model thousands of labeled pet photos, and each wrong guess slightly tunes it until it rarely misses.