Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Generative AI Leader practice bank has 2,191 more like it. Try 10 free: instant score, no account.
An enterprise customer service team uses a generative AI model to process warranty claims from customer emails. When incoming emails contain typical warranty requests with complete information, the model performs reliably. However, when an email has missing invoice numbers, unusual product configurations, or unsupported country codes, the model fabricates plausible-sounding receipt numbers and makes incorrect assertions about warranty coverage.
Which prompt design issue is the primary cause of this failure in edge cases, and what is the best way to address it?
A knowledge cutoff refers to the fixed temporal boundary of a model's pre-training corpus. Any historical events, product updates, or real-world occurrences that took place after this date are unknown to the base model.
The warranty assistant failed because it was asked about a brand-new product line released after the model's training date and had no external search or knowledge base connection.
Retraining large foundation models on daily operational data is cost-prohibitive and ineffective for handling missing customer inputs.
Few-shot prompting provides the model with one or more concrete input-output demonstrations within the prompt to illustrate desired formatting, structure, and reasoning styles before presenting the actual query.
[Missing ID] in the target output).The prompt contained contradictory, irrelevant, or logically inconsistent few-shot examples that actively misled the model during standard execution.
To fix edge-case hallucinations, few-shot examples should include demonstrations of edge cases and missing data handling rather than being removed entirely.
An underspecified task occurs when system or user prompt instructions fail to establish clear operating rules for non-standard inputs, missing information, or complex boundary scenarios. Because foundation models operate on probabilistic pattern matching, they naturally attempt to fill information voids with plausible-sounding content—leading to hallucinations when forced to process incomplete or rare edge cases without guardrails.
When a model lacks instructions for unexpected inputs, it defaults to predicting likely tokens, producing confident yet fictitious data. Providing comprehensive guidelines for missing data directly targets the root cause of edge-case failures without requiring expensive model retraining or complex architectural modifications.
Overt manipulation in prompt design involves using emotional appeals, flattery, or artificial pressure (such as writing "this is critical for my job" or "bad things will happen if you fail") in an attempt to improve response quality.
The prompt included extreme emotional pleas and urgency markers that caused degraded reasoning performance across all customer requests.
Clean prompt design should avoid artificial pressure, but robust systems still require explicit instructions for handling edge cases.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated