The opening chapter establishes what a generative model actually is before anything is asked of it. It covers foundation models and how they differ from a model trained for one labelled task, how text is tokenised and predicted one piece at a time, what a context window is and what happens when it fills, and why the same prompt sent twice can come back with two different answers. It also sets out the limits that follow from that design: a knowledge cut-off, no internal notion of truth, and confident phrasing that carries no guarantee behind it.
On the exam this chapter is worth seven of the 40 questions and seven of the 46 points. Only one of those is a K1 recall question; the other six are K2, so the examiner is not asking you to name a model family but to explain a behaviour or tell two concepts apart. Questions tend to present a short situation — a model that invented a plausible API method, an answer that changed after a version upgrade — and ask which property of generative AI explains it.
The most productive thing to carry into the exam from here is the separation between fluency and correctness. A tester who has internalised that a model optimises for a likely continuation rather than a true statement will answer most of this chapter correctly, and will also find the risk chapter easier later. Practise on the questions in this pool until the vocabulary — foundation model, prompt, token, context window, temperature, hallucination — is something you can define without reaching for a phrase from the syllabus.
This chapter is covered most heavily (7 questions) in CT-GenAI — Mock Exam 1 (Prompting for Test Analysis, Test Data and Pilot Governance), CT-GenAI — Mock Exam 2 (Model Behaviour, Technique-Driven Prompting and Operational Risk), CT-GenAI — Mock Exam 3 (Checkable Output, Retrieval Operations and Governance)