Page 55 - CyberSecurityEssentialsEnglish

Practical Exercises

↻ Exercise 1: Map your day
Goal: See AI as a set of task types rather than as one tool.
Required: Five tasks that recur in your week. Allow about 20 minutes.
Steps: Name each task with a verb. Identify its output type from the seven task families.
Decide whether AI would accelerate, augment, or automate it. Then write one sentence
on what an error would cost.
Output: A table of five tasks with the output type, the productive role, and the impact of
an error.
Assessment criterion: Each row names an output type, not a product. Naming a tool is

not an answer.

↻ Exercise 2: System or model?
Goal: Separate the model from the system built around it.
Required: One AI tool you already use. Allow about 20 minutes.
Steps: Name the model if the provider states it. Then list the interface, the data the tool

can reach, the tools it can call, and the rules that constrain it. Mark which of the five you
cannot see from outside.
Output: A five-part description of one tool, with the parts you cannot inspect marked.
Assessment criterion: The answer distinguishes what the provider documents from what
you inferred.

↻ Exercise 3: Fit test

Goal: Replace a general impression with an acceptance criterion you can check.
Required: One task from Exercise 1. Allow about 30 minutes.
Steps: Write ten cases, seven ordinary and three difficult, including one incomplete input
and one edge case. Write the acceptance criterion before you run anything. Run the
cases. Then count how many pass.
Output: Ten test cases, a criterion stated in advance, and a pass count.
Assessment criterion: The criterion was written before testing and another person could

apply it. A criterion such as the answer is good does not pass.

↻ Exercise 4: Next-unit experiment
Goal: Understand that fluency is a property of probability, not of truth.
Required: Three incomplete sentences from your field. Allow about 15 minutes.
Steps: For each sentence write five completions you expect a model to treat as probable.
Mark each one true, false, or not checkable. Then explain in two lines why the most

probable completion need not be the most truthful.
Output: Three sentences with five completions each, marked for truth.
Assessment criterion: The explanation refers to how the next unit is selected, not to the
system trying to be correct.

54