Training versus using a model

Training is the slow, expensive part, where the weights are adjusted on a huge dataset, sometimes for weeks on thousands of chips. Using a model (inference) is quick: the weights stay fixed and the model just turns an input into an output. A chatbot isn't retraining itself as you type; it applies what it learned to the text in front of it. Models are also tested on examples they never saw, to check they learned patterns rather than memorising.

Why it matters: Knowing which phase you're in explains costs, limits and why a model doesn't simply learn from each chat.

Builds on: Learning from examples

Learn it, then prove it

Sign in (free) to take this step's mastery check and save your journey.