Question 160 A content assistant gives different answers to the same low-risk extraction prompt, causing downstream churn. Which parameter should be tested? Increase temperature to maximize creativity Always select a different model randomly Lower generation randomness while retaining factual validation Remove output schemas from the API Explanation Correct Answer C:Reduced sampling variability can improve output consistency for constrained tasks. Incorrect Answers A:More variability can worsen repeatability of extracted records. B:Random routing introduces further uncontrolled variation. D:Without structure, the integration becomes harder to validate.
Question 161 A Python service sometimes sends requests to the wrong region. How should model endpoint configuration be managed? Use the first endpoint returned by an unrelated search Guess the region from the user's language Validate region and deployment references as explicit environment settings Select a new region per request based on model creativity Explanation Correct Answer C:A deterministic target prevents silent cross-region routing and compliance drift. Incorrect Answers A:Search order is not a release configuration contract. B:User locale does not authorize service residency choices. D:Nondeterministic location changes hinder compliance and observability.
Question 162 A contract RAG app answers questions from retrieved clauses without showing provenance. What should be added? Append a random legal website link to every sentence Persist document IDs and quoted spans with each supported answer claim Use one generic citation to the entire corpus Return only similarity scores with no excerpts Explanation Correct Answer B:Source links let a reviewer inspect the actual contract passage. Incorrect Answers A:Unrelated sites do not support specific contract obligations. C:Corpus-level references do not identify which clause justifies an answer. D:Numerical rankings cannot reveal the underlying contractual evidence.
Question 163 A contract scanner must recover handwritten annotations in a two-column PDF. Which TWO extraction features matter? Layout/reading-order analysis linking notes to nearby clauses Ignore annotations whenever digital text is present OCR that supports the source handwriting and printed text Generate typical legal annotations from prior examples Use file name only for contractual terms Explanation Correct Answers A:Spatial structure is necessary to interpret annotations in context. C:Handwritten amendments may carry obligations absent from embedded text. Incorrect Answers B:Printed text can be modified by handwritten exceptions. D:Common phrases are not this contract's source evidence. E:Names cannot recover document obligations.
Question 164 An AI classifier finds a phone number and labels it as a payment-card number. What validation helps? Use a larger text output to make the label seem certain Classify every digit sequence as a card Contextual type checks and format rules for sensitive entities Remove the digits without reporting classification uncertainty Explanation Correct Answer C:Similar digit strings require evidence beyond mere length. Incorrect Answers A:Verbosity cannot establish a correct entity type. B:False positives will misroute legitimate contact information. D:Hidden errors impair review and customer support.
Question 165 A service streams model tokens into a parser that expects one complete structured object. What should the client do? Treat network interruption as successful completion Commit every incomplete JSON fragment to production Buffer and validate complete structured output before committing records Guess missing brackets and fields whenever streaming pauses Explanation Correct Answer C:Partial streaming fragments are not complete typed business objects. Incorrect Answers A:Truncated output must not be reported as a verified record. B:Half-objects can corrupt downstream records. D:Synthetic repairs risk changing factual content and types.
Question 166 A model must distinguish a product defect complaint from a refund request. What classification behavior should it support? Ignore the user's verbs and classify by product name Always label any complaint as refund authorization Multi-intent detection or a documented primary-intent precedence rule Route every mixed message to product engineering only Explanation Correct Answer C:A message may mention both dissatisfaction and a requested financial action. Incorrect Answers A:Entity identity alone cannot determine requested action. B:Negative sentiment does not imply approval to issue money. D:An explicit refund request may require separate handling.
Question 167 A customer agent gets an error response with `isError=true` and human-readable text. How should it interpret the payload? Report the requested update complete because a response arrived As a tool failure requiring recovery or escalation, not verified business data Use it as permission to alter customer records Treat grammatical text as a successful search result Explanation Correct Answer B:Explicit error status is part of the tool result contract. Incorrect Answers A:Response receipt does not imply a successful action. C:An error cannot expand transaction authority. D:Readable error messages can resemble ordinary content.
Question 168 A team sees many tool calls for simple greetings from a forced tool-choice setting. What adjustment fits? Remove model text output entirely Force a data-write tool for all user turns Increase tool concurrency for greetings Permit direct text responses where no external data or action is needed Explanation Correct Answer D:Tool choice should reflect actual task requirements rather than mandatory use. Incorrect Answers A:The user still needs a normal conversational reply. B:Unnecessary external mutation is costly and unsafe. C:More unnecessary calls amplify the same routing flaw.
Question 169 A video model's description claims a person said a phrase that is absent from the transcript. How should it respond? Separate observed visual action from verified spoken transcript evidence Rewrite the transcript until it matches the description Treat any plausible quotation as recorded dialogue Delete the audio track and claim the quote confirmed Explanation Correct Answer A:Speech claims require audio/transcript support, not inferred lip movements alone. Incorrect Answers B:Changing evidence to fit inference misrepresents the source. C:A guess cannot be cited as actual speech. D:Missing audio makes the quotation less verifiable.
Question 170 A model deployment is recreated with different parameters, but the app keeps a cached old endpoint identifier. What should be tested? Configuration refresh and actual inference against the current deployment Only the old cached inference response rather than a new deployment test Only whether yesterday's deployment succeeded Only the number of search index documents Explanation Correct Answer A:Verification must use the new resource binding rather than assuming cache invalidation. Incorrect Answers B:Previous outputs do not verify the application now targets the replacement deployment. C:Old results are not evidence for a replacement deployment. D:Index counts cannot detect a stale deployment ID.