Question 300
A summary must be written in Spanish but retain exact English API field names for implementation. What should the prompt specify?
- Write the complete response in English to preserve identifiers.
- Replace all API fields with generic placeholders.
- Translate explanatory prose while preserving designated code identifiers and their original spelling.
- Translate every identifier into Spanish, including the literal JSON keys.
Explanation
Correct Answer
C:Separating prose language from literal technical tokens satisfies both audience and implementation requirements. Literal identifiers are part of the software interface and often must be copied exactly. Surrounding Spanish prose can explain them without renaming keys or changing example code.
Incorrect Answers
A:The user requested a Spanish explanation, which can coexist with unchanged technical names.
B:Placeholders remove the exact references needed to implement the guidance.
D:Changing implementation identifiers can make the resulting instructions unusable against the actual API.
Question 301
A message says, 'I love the phone, but the delivery service was terrible.' The task is to classify sentiment about delivery. What instruction is essential?
- Average sentiment over all entities without recording which one was requested.
- Evaluate sentiment toward the named target rather than collapsing all mentioned entities.
- Use the first sentiment-bearing phrase as the overall label.
- Treat any positive word as a positive delivery rating.
Explanation
Correct Answer
B:Target-specific classification preserves the negative delivery judgment despite positive sentiment elsewhere. The positive phone assessment and negative delivery assessment have different targets. The model should associate the sentiment expression with the requested entity before assigning the label.
Incorrect Answers
A:An overall average would answer a different question from the target-specific task.
C:The first phrase concerns the phone, not the specified delivery target.
D:A positive word about another entity does not establish sentiment toward delivery.
Question 302
A source reports that an error rate rose from 10% to 12%. A summary must distinguish the absolute percentage-point change from the relative percentage increase. Which wording is accurate?
- The error rate rose by 2 percentage points, a 20% increase relative to its original rate.
- The error rate rose by 12 percentage points because 12% is the final reported rate.
- The error rate rose by 2%, and the two measures can be used interchangeably.
- The error rate rose by 20 percentage points, which is a 2% relative increase.
Explanation
Correct Answer
A:Subtracting 10% from 12% gives the percentage-point change. The relative increase uses the original rate as the denominator, so 2 divided by 10 equals 20%.
Incorrect Answers
B:The endpoint is not the amount of change. The absolute change must account for the starting rate as well as the final value.
C:A bare 2% increase normally describes a relative change and obscures the percentage-point distinction requested. The summary must label which calculation it reports.
D:This swaps the two measures. The rate difference is 2 percentage points; dividing that difference by the original 10% gives a 20% relative increase.
Question 303
A request to 'summarize the report' omits a preferred tone, but names the audience and purpose clearly. No decision depends on tone. What should the prompt policy favor?
- Stop for clarification about every possible stylistic preference.
- Use a reasonable tone suited to the stated audience and complete the summary.
- Invent a new audience so all parameters are explicit.
- Transcribe the report verbatim to avoid making any stylistic judgment.
- Refuse to summarize any document without a fully specified style guide.
Explanation
Correct Answer
B:A reversible stylistic assumption allows progress when it does not affect the task's substantive correctness. Clarification is valuable when an ambiguity changes correctness, scope, or a consequential decision. A low-impact tone preference does not justify blocking an otherwise well-defined summarization task.
Incorrect Answers
A:Low-impact optional details need not block a useful summary when the objective is clear.
C:Replacing a supplied audience disregards the user's actual request.
D:A transcription does not perform the requested summarization. The specified audience and purpose provide enough guidance to choose a reasonable tone while selecting the important content.
E:A complete style guide is not necessary for an ordinary audience-specific summary.
Question 304
A product-comparison prompt asks, 'Explain why our product is always superior,' but the task is an impartial evidence review. What should change?
- Ask the model to omit all competitor strengths.
- Ask for comparison against explicit criteria, including strengths, weaknesses, and missing evidence.
- Require a positive conclusion but let the model choose which supporting facts to mention.
- Keep the conclusion fixed but ask for more citations.
Explanation
Correct Answer
B:A criteria-based prompt allows conclusions to follow the supplied evidence rather than a predetermined winner. The rubric should let unfavorable findings survive when evidence supports them. Gaps in comparative data should remain explicit rather than being filled with promotional assumptions.
Incorrect Answers
A:Selective evidence removal conflicts with the stated impartial-review objective.
C:A fixed favorable outcome still biases selection away from an impartial evidence comparison.
D:Citations do not make a predetermined unsupported conclusion impartial.
Question 305
A generated description calls a service 'the fastest in the industry,' but the supplied material contains no comparative benchmark. What is the appropriate edit?
- Change fastest to best while preserving the unsupported ranking.
- Remove or qualify the superlative and retain only supported performance claims.
- Cite the service's own feature list as proof of industry leadership.
- Prefix the unsupported ranking with 'we believe' while presenting it as a verified comparative fact.
Explanation
Correct Answer
B:An industry-wide ranking requires comparative evidence that the supplied material does not provide. A feature list can support claims about available functionality, but not a market-wide speed ranking. A defensible comparative claim needs a defined benchmark and relevant comparison set.
Incorrect Answers
A:A different superlative retains the same evidence problem.
C:A feature list without comparative measurements does not substantiate fastest-in-industry performance.
D:A belief phrase does not supply the missing benchmark. The description should distinguish opinion from any evidence-backed performance claim.
Question 306
An extraction test expects 'United States,' while correct outputs may say 'USA.' Which TWO evaluation choices avoid falsely penalizing equivalent answers? Choose TWO.
- Require every correct answer to copy the reference's punctuation exactly.
- Accept every fuzzy string match above an arbitrary threshold without checking whether it names the same country.
- Map every country value absent from the approved alias list to United States by default.
- Check that normalization preserves distinctions important to the task.
- Define an approved normalization or equivalence mapping for the field.
Explanation
Correct Answers
D:Equivalence rules must not collapse meaningfully different values merely to increase scores.
E:An explicit mapping can recognize synonymous representations without accepting unrelated values.
Incorrect Answers
A:Incidental formatting should not create failures when the task permits equivalent names.
B:Similar spelling can identify different countries. Equivalence should preserve the field's meaning rather than rely on an unchecked surface-distance rule.
C:An unknown value is not evidence for the reference country. This default would turn unsupported or different countries into false matches.
Question 307
A prompt demands a 20-word summary containing 15 detailed findings. Both requirements cannot reasonably be met. What should the designer do?
- Set an explicit priority or revise the length/content requirements to a feasible contract.
- Let the model omit findings without defining which requirements may be sacrificed.
- Keep both constraints and ask the model to compress each detailed finding into one word.
- Penalize the model for every output without changing the requirements.
Explanation
Correct Answer
A:A workable constraint hierarchy tells the model what to preserve when requirements compete. For example, the contract could prioritize all findings with a larger limit or request a short overview plus a separate detailed list. The designer must decide which deliverable is actually required.
Incorrect Answers
B:Undocumented omissions make compliance unpredictable when length and completeness compete.
C:Extreme compression does not preserve the detail required by the task and leaves the contract infeasible.
D:An infeasible contract cannot be repaired by harsher evaluation alone.
Question 308
A prompt starts, 'You are a world-leading accountant.' What can this role instruction reasonably help with?
- Framing terminology and style, while factual and numerical correctness still require evidence and checks.
- Making all generated tax figures correct without source data.
- Eliminating the need to define the actual accounting task.
- Treating the role label as a substitute for providing the jurisdiction and rules relevant to the task.
Explanation
Correct Answer
A:A role can guide presentation but does not confer credentials or guarantee accurate accounting results. Useful role framing does not replace the source figures, applicable rules, or validation required by the task. Evaluation should measure the resulting work rather than the prestige of the assigned persona.
Incorrect Answers
B:Expert framing cannot supply missing facts or validate arithmetic.
C:A persona does not specify the inputs, output, or acceptance criteria of the requested work.
D:A role description cannot supply missing task-specific constraints. Correct treatment still depends on the applicable evidence and requirements.
Question 309
An extraction task requires finding invoice lines, computing totals, and writing a short explanation. How can the prompt make errors easier to diagnose?
- Merge all source amounts into a single unlabelled list.
- Separate extracted line items, validated totals, and the final explanation in the task contract.
- Request only a final persuasive paragraph with no structured intermediate outputs.
- Combine the final total and explanation in one free-text field while omitting the extracted line values.
Explanation
Correct Answer
B:Distinct observable outputs let the application localize discrepancies without relying on hidden reasoning. These are observable task products, not a demand for hidden internal reasoning. The application can compare extracted values with the source and recompute totals before accepting the narrative.
Incorrect Answers
A:Removing line identities makes reconciliation harder rather than easier.
C:A prose-only result obscures whether an error arose in extraction or calculation.
D:This hides the intermediate data needed to distinguish an extraction error from a calculation error. Separate outputs make the failing stage observable.
Question 310
A policy says, 'If the customer opts in, send a reminder.' The customer has not opted in. What should an action extractor preserve?
- A conclusion that reminders are always forbidden.
- An inference that reading the policy counts as opting in.
- The reminder action together with its unsatisfied opt-in precondition.
- An unconditional instruction to send the reminder now.
Explanation
Correct Answer
C:Conditional extraction must retain the trigger that determines when the action becomes applicable. The extractor can represent an action as conditional without scheduling it immediately. Losing the opt-in condition would turn a description of policy into an unauthorized operational instruction.
Incorrect Answers
A:The policy permits reminders when the stated condition is satisfied.
B:The source gives no basis for treating policy access as customer consent.
D:Removing the condition changes the policy into an action it does not currently authorize.