AWS Certified AI Practitioner (AIF-C01) Practice Questions

AWS Certified AI Practitioner (AIF-C01) exam preparation

AWS Certified AI Practitioner (AIF-C01) practice questions in English: 180 questions with explanations. The mock exam has 65 questions and 90 minutes.

AWS Certified AI Practitioner (AIF-C01) free question 3 / 10 · Domain 2: Fundamentals of GenAI

When asked about internal policies, a chatbot sometimes confidently cites clauses that do not exist in the policies. Which two methods does the Amazon Bedrock "Prompt engineering concepts" page list for reducing hallucinations? (Choose two.)

  1. Buy Provisioned Throughput to raise the baseline quality of responses
  2. Switch to a different model and check whether it gives better results
  3. Count the input with the CountTokens API and trim it so it does not exceed the limit
  4. Use RAG so the model can refer to relevant data
  5. Raise the temperature inference parameter to widen the candidates for output
Show the answer and explanation

Answer: B, D

Why this is correct
A Note on the page says that to reduce hallucinations you can refine the prompt with prompt optimization techniques, use a technique such as RAG so the model can reference relevant data, or use a different model that may give better results. For questions whose answers are in documents you hold, such as policies, passing the relevant part of the policy to the model with RAG is effective.

Why the other options are wrong
"Buy Provisioned Throughput to raise the baseline quality of responses" … Provisioned Throughput is a purchasing mechanism that secures the volume and speed of input and output a model processes for a fixed cost. It does not make responses more correct.
"Raise the temperature inference parameter to widen the candidates for output" … The inference parameters page says raising the temperature makes lower-probability outputs more likely to be chosen, so responses become more random. That is not a way to reduce answers that do not match the facts.
"Count the input with the CountTokens API and trim it so it does not exceed the limit" … CountTokens is an API for estimating cost before inference or adjusting to fit token limits (see the page on counting tokens). It is not listed as a way to reduce hallucinations.

Key point
The Generative AI Lens defines hallucination as the phenomenon where a generative AI model produces output that is inconsistent, factually incorrect, or irrelevant to the input prompt. The same list of definitions says RAG is used to ground responses in documents and reduce hallucinations.

Go to the AWS Certified AI Practitioner (AIF-C01) practice app (10 free questions; purchase unlocks all questions)

See the full practice set