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.
A global enterprise deploys a foundation large language model (LLM) to assist compliance officers with policy research and regulatory reporting. During initial user testing, the compliance team identifies two critical failure modes:
Which pair of foundation model limitations directly explains these behaviors?
Cold start latency refers to the delay experienced when an AI model endpoint or serverless container initializes from an inactive state before it can serve its first request. Prompt injection is a security vulnerability where untrusted user input manipulates the model into bypassing its system instructions, safety guardrails, or intended behavior.
The knowledge cutoff refers to the fixed point in time at which a foundation model's training dataset was finalized, creating a temporal boundary beyond which the model has no innate awareness of new events or updated facts. Data dependency describes the foundational principle that a model's internal understanding, factual accuracy, and reasoning patterns are strictly determined by the quality, scope, and integrity of the datasets on which it was trained.
This option directly identifies the dual root causes of the compliance tool's issues: temporal blind spots caused by the static nature of pre-training sets (knowledge cutoff) and reliance on the quality of underlying historical training data (data dependency).
A token context window defines the maximum number of input and output tokens an LLM can process in a single request or conversation session. Temperature scaling is an inference hyperparameter that controls the randomness of token selection during generation, where higher values yield more creative outputs and lower values yield more deterministic outputs.
Model drift (concept drift or data drift) occurs when the statistical properties of production input data change over time relative to the baseline training data, leading to degraded predictive performance in traditional machine learning. Catastrophic forgetting occurs when an already trained neural network completely loses previously learned skills or knowledge upon being fine-tuned or trained on a new, distinct dataset.
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