The process of teaching a model by exposing it to data and adjusting its parameters to minimize errors.
Friendly Description: Training is the long process of teaching an AI model how to do its job. The model is shown lots of examples, makes guesses, gets corrected, and slowly improves. Training is the part that takes the most time, computer power, and money, but once it's done, the trained model can be used over and over to do useful work.
Example: Training a modern language model can take weeks of running on thousands of specialized chips, processing trillions of words. Once that one-time investment is made, the resulting model can answer billions of user questions over its lifetime, like building a great chef in school so they can cook meals for years.