Learn neighboring concepts together
Compare terms that sound similar: precision and recall; prefill and decoding; fine-tuning and retrieval; explainability and interpretability. Write a concrete example and a non-example for each. Knowing a term means knowing what it does not establish.
Connect vocabulary to decisions
For every term, ask what it changes in a real system. A KV cache affects inference; an embedding supports representation; an approval binding controls a particular action. Avoid substituting a technical label for evidence that a system works.
Recognize the limits of a label
Some terms have precise definitions while others vary by author or organization. AGI and frontier model are not universally fixed thresholds. Open weights does not automatically mean unrestricted licensing. Read the surrounding assumptions before selecting an answer.
Filter by subject, then try explaining each term without reading its definition.
A worked example
Neither guarantee follows. Retrieval can supply relevant evidence but generation can still make unsupported claims. A context window is a capacity limit, not proof that every supplied fact will be used correctly.
The point: Translate the jargon into a claim you can test.
Before you take the exam
Explain these concepts in your own words, work through a different example, and describe why the tempting alternatives fail. Use the references to go deeper.