Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 41

Question 400

An app pays to generate identical static onboarding text repeatedly. What should be measured before changing inference spending?

Cache personalized account status for every user under one shared onboarding key.
Precompute approved immutable responses while retaining dynamic-policy checks.
Regenerate identical permanent boilerplate on every request despite no variable input.
Present precomputed content forever without revalidation when source policy changes.

Explanation

Correct Answer

Precompute approved immutable onboarding copy, while validating current policy exceptions and avoiding stale dynamic state in reused responses.

Incorrect Answers

Static template reuse must not leak private or mutable account data.

Unnecessary model calls inflate cost and latency.

A static answer must be invalidated when its policy basis changes.

Question 401

A caching approach stores protected model outputs without encryption. Which optimization balances actual cost and accepted task quality?

Apply cache access control, encryption and retention appropriate to content.
Store private answers in an unencrypted global public cache because they're already generated.
Treat a short cache TTL as equivalent to access control while records remain readable during TTL.
Rely on a prompt instruction not to reveal cached secrets rather than restrict cache readers.

Explanation

Correct Answer

Protect cached model outputs with identity-scoped access, encryption, lifecycle/expiry and sensitivity-aware retention.

Incorrect Answers

Generation does not declassify confidential content.

Expiry cannot stop unauthorized reads before deletion.

A language model cannot enforce backend cache permissions.

Question 402

A team pays for Provisioned Throughput but wants to know whether moving to a different supported model/deployment tier would lower the cost of ACCEPTED business answers without degrading accuracy. Which TWO evaluation measurements are necessary?

Rank deployment choices only by their advertised hourly capacity price, ignoring accepted outputs.
Scale provisioned capacity until queue latency falls and assume quality automatically improves.
Calculate total workload spending divided by business-accepted correct answers, including remediation when measured.
Use a public model parameter-count leaderboard as the only proxy for answer correctness.
Compare supported tiers on the same representative workload, quality thresholds, end-to-end performance and resulting unit cost.

Explanation

Correct Answer

Accepted-answer unit cost links inference charges and actual business outcomes rather than just request volume.

A same-workload comparison can decide whether a cheaper or faster tier meets required quality and response-time constraints.

Incorrect Answers

Nominal price cannot establish cost per useful task if models differ in token demand and failure rate.

Extra provisioned resources may reduce queuing but do not fix unsupported or incorrect generated content.

Model size does not reliably represent domain-specific answer quality or whether the result passed business checks.

Question 403

A cost review counts only input tokens. What should be measured before changing inference spending?

Account for output and applicable cache/write/batch token charges.
Estimate cost from input tokens alone when responses are long.
Count every cached input token at the same rate as an uncached inference token.
Use a stale price table for a different model SKU as the cost basis.

Explanation

Correct Answer

Account for all applicable input, output, cache read/write and batch usage meters under the model's actual pricing schedule.

Incorrect Answers

Output token charges can dominate model spend.

Cache reads and writes may use different billing rates.

Published prices vary by exact model and operation.

Question 404

A nightly inference job misses budget due to oversized prompts. Which optimization balances actual cost and accepted task quality?

Combine context selection, output cap and scheduled batch where supported.
Move an interactive account-update workflow into asynchronous batch to get a discount.
Remove safety and business constraints indiscriminately to shorten prompts.
Increase output-token limits for every night job without checking whether summaries need them.

Explanation

Correct Answer

Reduce irrelevant input context and unnecessary output length; use scheduled batch only for supported noninteractive work with source-complete answers.

Incorrect Answers

Batch lacks the required live tool interaction/transaction controls.

Token reduction cannot justify unsupported or unauthorized outcomes.

Oversized completion budgets can worsen spend.

Question 405

A team tunes costs using stale pricing documentation. Which resource-efficiency decision fits the workload?

Reuse an old model provider price slide without checking current SKU and region.
Treat per-call price as constant regardless of input/output tokens and cache behavior.
Commit provisioned capacity solely from a discounted list price without utilization evidence.
Verify current model-SKU prices and quota terms before cost commitments.

Explanation

Correct Answer

Recheck the exact model SKU's current regional token rates, cache/batch rates and quota limits before making financial commitments.

Incorrect Answers

Pricing and availability change over time.

Different workload shapes and operations can change charges.

A lower unit rate can still cost more with idle committed capacity.

Question 406

A streaming chat takes too long to show first text. What improves this path without silently weakening answer quality?

Simulate streaming by typing out the fully generated text after a nonstreaming call returns.
Use supported stream path and measure time-to-first-token separately from full completion.
Optimize only S3 bucket writes and ignore the model's supported streaming operation.
Treat the first UI loading spinner as a model token for latency measurement.

Explanation

Correct Answer

Use a supported streaming model/API and measure real time-to-first-token and end-to-end completion separately from UI placeholder timing.

Incorrect Answers

An animation does not reduce real time until the first token is available.

Object storage throughput is not necessarily the chat response bottleneck.

A spinner is not generated output and distorts the TTFT metric.

Question 407

A user receives first token fast but whole response misses deadline. Which measurement or tuning step should happen first?

Report only the first token's latency as the entire response-time performance.
Limit analysis to mean latency and omit slow tail cases.
Mark a partial streamed answer complete before its final stop/error event.
Track end-to-end completion and p95/p99, not initial token alone.

Explanation

Correct Answer

Track full successful response completion and p95/p99 latency, not merely early token arrival; long generation and tools can dominate SLA.

Incorrect Answers

Usable answer completion can arrive much later.

The p95/p99 SLA may fail even when average performance looks fine.

A partial response cannot satisfy the full completion deadline.

Question 408

An application invokes several independent retrieval tools sequentially. Which performance correction follows from the described bottleneck?

Parallelize safe independent lookups and reconcile results after all complete.
Parallelize dependent financial writes as if they were independent search requests.
Run every lookup serially even when none depends on earlier output.
Return the first finished tool's answer without waiting for other required evidence.

Explanation

Correct Answer

Run authorized independent read-only lookups concurrently, bounded by quotas, then reconcile all results and errors before generation.

Incorrect Answers

Out-of-order side effects can corrupt business state.

Unnecessary serial execution adds the individual wait times.

Missing sources can make the final response incomplete.

Question 409

A synchronous API grows its latency because reranking uses hundreds of hits. What improves this path without silently weakening answer quality?

Rerank all retrieved documents on every request regardless of candidate size.
Bound candidates before rerank and measure recall/latency tradeoff.
Remove reranking entirely without testing whether answer precision drops.
Increase the FM output budget as the remedy for upstream reranking latency.

Explanation

Correct Answer

Bound candidate count before expensive reranking, comparing relevance/recall and tail latency using representative queries.

Incorrect Answers

Unbounded rerank work causes latency and cost.

A speed-only change can sacrifice source quality.

Generation tokens cannot eliminate the slow retrieval stage.

Question 410

An FM gets slower as conversation history grows. Which measurement or tuning step should happen first?

Send the entire multi-year conversation to each FM call indefinitely.
Prune/summarize context using supported retained state without losing key facts.
Delete all earlier session facts without keeping necessary case state.
Use one globally shared summary for unrelated actors' conversations.

Explanation

Correct Answer

Prune irrelevant history and persist authorized summaries/critical state outside the prompt, measuring quality while reducing context length.

Incorrect Answers

Long histories increase tokens and eventually exceed the context window.

Missing essential details can create inconsistent answers or tool mistakes.

Cross-user state pooling is a privacy and correctness defect.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 400

Explanation

Correct Answer

Incorrect Answers

Question 401

Explanation

Correct Answer

Incorrect Answers

Question 402

Explanation

Correct Answer

Incorrect Answers

Question 403

Explanation

Correct Answer

Incorrect Answers

Question 404

Explanation

Correct Answer

Incorrect Answers

Question 405

Explanation

Correct Answer

Incorrect Answers

Question 406

Explanation

Correct Answer

Incorrect Answers

Question 407

Explanation

Correct Answer

Incorrect Answers

Question 408

Explanation

Correct Answer

Incorrect Answers

Question 409

Explanation

Correct Answer

Incorrect Answers

Question 410

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests