Training vs. Using: Where the Knowledge Comes From

There are two very different phases in an AI model's life. First comes training: the model is fed enormous amounts of text, images, and code, and slowly adjusts billions of internal settings until its predictions get good. This takes huge amounts of computing power and happens once, ahead of time, at the company that builds the model.

Second comes use, which is what you do. When you type a prompt, nothing is being learned or saved into the model. It is applying the patterns it already has, frozen at the point training stopped.

That is why a model can be out of date on recent events, and why it does not remember you between separate chats unless the product adds a memory feature on purpose. It is also why the same underlying model can power many different apps: the training is shared, the interface is different.

Takeaway: the model's knowledge has a cutoff. For anything recent or specific to you, you have to put that information into the conversation yourself.