Training and inference
Training adjusts a model using many examples.
Lesson 1 of 4 · Learn it by trying itWatch: How AI works without the technical headache
An original animated overview. No sound—follow the on-screen explanation, or read the transcript.
Read the video transcript
Training and inference. Training adjusts a model using many examples.
Tokens and context. Models process chunks called tokens, which may be words or parts of words.
Probabilities and variation. A language model predicts plausible token sequences.
Tools and grounding. Search and document retrieval can supply relevant evidence, but the model can still misread it.
Created for AI For Numpties · Original course material · Hosted with this website
Getting your practice ready…
Explore the idea with an example
Training adjusts a model using many examples. Inference is using that trained model to produce a response. An ordinary conversation does not necessarily update model weights; retention and future training depend on the provider’s policy and settings.
Think of training as adjusting a recipe after many practice batches, and inference as baking with the current recipe. This is an analogy, not a description of a model thinking. Changing the ingredients supplied in a prompt changes that response without necessarily changing the trained model. A provider may separately retain conversations or use them for later training. Read its current controls before entering information; “the model does not learn immediately” does not mean “the service stores nothing”.