An AI model is a system that has learned patterns from examples. A language model uses those patterns to produce text in response to your request. The website or app you chat with adds the interface and may add search, files, or other tools.
Learning first. Answering later.
Training adjusts the model’s internal numbers using examples. Inference means using the trained model. When you download a local model and ask it a question, you are normally doing inference. You do not need to train a model yourself to use one.
What does 14B mean?
The B stands for billion parameters: the learned numbers inside the model. They are saved as model weights. Size influences memory needs, but a larger number is not a guarantee of better answers. The examples used for training and the design of the model matter too.
A chat is not automatically a lesson
Giving a model your name lets it use your name in the current conversation. It does not necessarily change its weights. An app may separately save preferences or use conversations in a later training process, depending on its settings and policies.
Try this
Ask an AI to explain the same idea first to a ten-year-old, then to a specialist. The model can change its answer using your instructions without being trained again.
Choose the model and the app together. The model’s abilities matter; so do the tools and information the app lets it use.
Sources & further reading
These references go into more detail; some are written for developers.