Answer all 12 scored questions, then submit. Questions 8–10 and 12 describe realistic situations rather than asking for a definition directly — this is deliberate: the goal is to check that you can apply what the handout covered, not just recite it. Question 13 is a short reflection and isn't scored, but is worth answering honestly before you move on.
Question 1
A large language model (LLM) generates its output primarily by —
Correct answer: predicting plausible next-word patterns. This is exactly why fluent, confident output is not the same thing as verified output.
Question 2
"Hallucination," as used in this course, refers to —
Correct answer: fabricated or wrong output delivered with the same fluent confidence as correct output — the core failure mode covered in Handout Section B.
Question 3
According to the handout, the moment an AI's failure modes (hallucination, overconfidence, bias) become most dangerous is when —
Correct answer: when the output confirms your existing suspicion. This is confirmation bias, and it's the specific trap Module 4 is built to train against.
Question 4
"Multimodal AI" describes a system that —
Correct answer: working across multiple input types (e.g. a clinical note plus an imaging study) in the same session.
Question 5
A diagnostic AI tool may under-perform on your patients mainly because —
Correct answer: performance depends on the training/validation population — "the tool works well" is never automatically true for a population it wasn't tested on.
Question 6
A tool that listens to a consultation and drafts a structured clinical note afterward belongs to which category?
Correct answer: ambient documentation — the subject of Module 2.
Question 7
A tool that flags a possibly abnormal chest X-ray for a radiologist's review belongs to which category?
Correct answer: diagnostic decision support — the subject of Module 3.
Question 8 · Applied
A colleague tells you: "The AI suggested this diagnosis, and it matched what I was already thinking, so I went with it." What is the most accurate way to describe the risk in what your colleague just did?
Correct answer: the third option. Agreement between AI output and prior belief feels like confirmation, but it is precisely the situation the handout flags as most dangerous.
Question 9 · Applied
You are evaluating an AI documentation tool for a clinic with intermittent internet access and a mostly Tetum- and Portuguese-speaking patient population. What should you check first, before cost or features?
Correct answer: connectivity tolerance and language fit come first — a feature-rich tool that can't function offline or doesn't handle your working languages isn't usable, however advanced it is.
Question 10
In the KNOW / FIND / THINK model, where does "the exact subscription price of a specific AI product" belong?
Correct answer: FIND. Specific, frequently-changing details like pricing don't belong in memory — know where to check them when the decision is real.
Question 11
True or false: an AI tool that states its answer confidently is, for that reason, more likely to be correct than one that hedges.
Correct answer: False. Confidence is a tone the software generates, not a measure of its own accuracy — this is the handout's core warning.
Question 12 · Applied
A patient arrives having already run their symptoms through an AI chatbot and hands you its printed conclusion. At this stage of the course (Module 1), what is the most appropriate stance?
Correct answer: the middle option. Module 1 only establishes the mindset; the practical skills for handling this exact situation are built in Module 3.
Question 13 · Reflection (not scored)
In your own words: why does this course insist on "use AI to extend your thinking, not replace it," rather than simply teaching you to use the tools?