noun · AI concept
Model
also called AI model
The brain without the body. A big file of numbers that turns text in into text out, and does nothing else on its own.
A trained program whose behavior comes from a huge set of learned numbers instead of rules someone typed, which maps a sequence of tokens in to a sequence of tokens out.
A trained neural network: a fixed set of numbers (weights) plus the code shape that uses them. It maps input tokens to a probability for the next token, and it keeps no state between calls.
Real-life example
You open a chat app and ask for a recipe. It feels like one thing is answering, but it is two: the app, which collects your message and shows the reply, and the model behind it, which only ever sees text and returns text. Swap the model and the app looks the same, but swap the app and the same model behaves differently.
Where the analogy breaks
-
The model is not the product. The apps you use add memory, tools, files and safety checks around a model. Most of what feels like "the AI" is that wrapper.
-
It does not learn from your chat. Its numbers are fixed when you use it. Whatever it seems to remember comes from text the app sends it again.
-
A bigger model is not always the better choice. Larger models usually answer harder questions better, but they cost more and respond slower, so small ones often win on simple jobs.