Level 1 exam — Generative AI at Work

Exam — fresh questions

12 questions to validate the art of prompting: RCTF template, iteration, few-shot and common mistakes.
Answer everything, then click Check: your score and explanations will appear.

1.1 · Recall

1. What are the 4 blocks of a professional prompt?

Role, Context, Task, Format Subject, verb, object, signature Beginning, middle, end, example Question, answer, summary, send

RCTF: role calibrates, context avoids assumptions, task orders, format frames the output.

1.1 · Application

2. Amine wants a schedule-change announcement. Which prompt is pro?

“You are a project manager [role]. Delivery postponed to the 30th, stakeholders to notify [context]. Write the email [task]. Subject + 120 words, reassuring tone [format]” “Write something nice about my project, thanks” “Write me an announcement for the project” and nothing else “Tell me about project delays in general”

All four blocks are there: role, context, precise-verb task, format with length and tone.

1.1 · Application

3. Amine's prompt returns an 800-word essay when he wanted 5 bullets. What was missing?

A leading question to guide the model Several questions asked at the same time The Format block (length, structure, tone) A second example identical to the first

Format is the highest-ROI block: without it, the AI chooses length and structure at random.

1.2 · Recall

4. What do effective users do with a first imperfect answer?

They copy the answer as it is without proofreading They retype the whole request from scratch They iterate with short instructions in the same thread They abandon the tool for this type of task

The thread keeps the context: “shorter”, “different tone”, “add X” sculpt the answer.

1.2 · Application

5. Amine's invitation is long and flat. Which sequence is correct?

Ask four new questions in the same message Copy the raw text with no touch-ups Throw away the thread and retype the same request six times “Shorter, 80 words max” then “More enthusiastic” then “Add the agenda in 3 bullets”

Three short iterations bring the answer close to final, then ask for variants.

1.2 · Arbitration

6. After 8 contradictory back-and-forths, Amine's thread gets tangled. Arbitration?

Restart a fresh thread with a consolidated prompt integrating everything learned Persist ten more messages on the same thread Definitively abandon the current task Send the latest version without proofreading it

Beyond 4-5 back-and-forths, the initial framing was bad: fresh thread + consolidated prompt.

1.3 · Recall

7. What is few-shot prompting?

Asking several questions in a single message Authoritatively limiting the length of answers Demanding an answer within a very short time Providing an example of the expected result to imitate

Showing beats describing: one example beats ten lines of explanations on structure, tone, length.

1.3 · Application

8. Amine wants five consistent progress sheets. Which method?

Describe the format in ten lines with no example Write ONE perfect sheet then: “On exactly this model, write the four others” Request the five sheets with no form guidance at all Stack eight different examples in the same prompt

One good benchmark is enough: the AI mirrors structure, tone and length with unmatched fidelity.

1.3 · Arbitration

9. Amine's benchmark contains client names and real amounts, on an unapproved tool. Arbitration?

Publish the example on a forum to get an opinion Anonymise with fictitious values: the model only needs the FORM Add even more real data to be precise Send as it is: real data improves imitation

Real data = needless risk: form alone suffices for imitation. Reflex detailed in 5.1.

1.4 · Application

10. Amine asks “Tell me about time management”. What is the problem and the fix?

Leading question: add “it's good, right?” at the end No problem: this is the ideal prompt Multiple request: ask even more questions at once Vague request: specify task, angle and use (“5 techniques for an overwhelmed manager, one sentence each”)

Vague → specify task, angle and use; otherwise the AI produces a lukewarm encyclopedia.

1.4 · Application

11. Amine crams four questions into a single message and gets superficial answers. Fix?

Ask for an even shorter answer covering all four Ask them one by one, one question per message Re-ask them all together but in capitals Add a fifth question to force effort

Multiple request → split: each question deserves its message and its full answer.

1.4 · Arbitration

12. Before the committee, Amine wants an honest opinion on his schedule. What should he ask?

Nothing: the AI's enthusiasm proves quality “Tell me only what is perfect in this plan” “Critique this schedule like a sceptical client: 3 major weaknesses” “My schedule is solid, right?” to be reassured

The model is sycophantic by default: only explicit critique gives analysis useful for decisions.

What's next?

Score ≥ 80%: move on to Level 2 V3 (write, summarise, translate, brainstorm and structure with AI day to day).
Score between 50% and 80%: review the failed sections (RCTF in 1.1, iteration in 1.2, few-shot in 1.3, mistakes in 1.4) via their explanations above, then retake the test.
Score < 50%: reread the level while practising each exercise (template, 4-step sculpting, consistent series, failed-prompt clinic) before retaking the exam.