Model training

During training, a model adjusts billions of numbers inside it so its predictions fit the data better. Its knowledge stops at the date of its training data, so it does not know later events unless you give it sources. Using a model in a chat does not retrain it directly, but providers may have their own policies on conversation data.