AI0-001 Practice Test 24 — 15 Questions

Question 1

A data scientist is preparing a dataset for a classification task. The dataset contains 10,000 rows and 50 features, but many features have missing values. Which approach should the scientist take first to address the missing data?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Analyze the pattern and proportion of missing values to choose an appropriate imputation strategy.. The first step in handling missing data is to understand the pattern and proportion of missingness (e.g., MCAR, MAR, MNAR) to select an appropriate imputation method. Blindly applying imputation or deletion without analysis can introduce bias or reduce model performance. This diagnostic step ensures the chosen strategy aligns with the data's underlying structure and the classification task's requirements.

Answer

Use a deep learning model to predict missing values without preprocessing.

Answer

Remove all rows with any missing values to ensure a clean dataset.

Answer

Replace missing values with the mean of each feature immediately.

Question 2

During an audit of an AI system, the auditor requests documentation on the model's intended use, performance metrics, and limitations. Which tool is designed to provide this information in a standardized format?

Accepted Answer

Model card. A model card is a standardized document that describes a model's intended use, performance metrics, limitations, ethical considerations, and other relevant details. It is specifically designed to provide transparency and accountability documentation for AI models, matching the auditor's request. Model cards were popularized by Google and are now a common governance artifact.

Answer

SHAP values

Answer

LIME

Answer

Data card

Question 3

An organization wants to fine-tune a 7B parameter LLM for a specialized legal document summarization task. They have a small labeled dataset (500 examples) and limited GPU budget. Which THREE techniques should they consider? (Choose three.)

Accepted Answer

Use LoRA (Low-Rank Adaptation). Option A (LoRA) is correct because Low-Rank Adaptation freezes the pretrained 7B weights and trains small low-rank adapter matrices, drastically reducing trainable parameters and GPU memory, which fits the limited GPU budget. Option B (instruction-tuning dataset with input-summary pairs) is correct because the 500 labeled examples must be formatted as supervised instruction/response pairs so the model learns the specific legal summarization mapping during fine-tuning. Option E (QLoRA with 4-bit quantization) is correct because QLoRA quantizes the frozen base model to 4-bit (e.g., NF4) and trains LoRA adapters, further cutting memory so a 7B model can be fine-tuned on modest GPUs. Option C is not appropriate because training a new model from scratch on legal text requires massive compute and data, far beyond 500 examples and a limited GPU budget. Option D is not appropriate because full fine-tuning updates all 7B parameters, demanding very high GPU memory and compute that the organization's budget cannot support.

Answer

Train a new model from scratch on legal text

Answer

Full fine-tuning of all model parameters

Question 4

A data scientist is building a model to predict whether a transaction is fraudulent. The dataset has 99.9% legitimate transactions and 0.1% fraudulent ones. Which evaluation metric is MOST appropriate to assess model performance given this class imbalance?

Accepted Answer

F1-score. With 99.9% legitimate transactions and only 0.1% fraudulent ones, accuracy would be misleadingly high (99.9%) even if the model never predicts fraud. The F1-score is the harmonic mean of precision and recall, making it robust to class imbalance by penalizing both false positives and false negatives. This makes it the most appropriate metric for evaluating fraud detection performance.

Answer

BLEU score

Answer

Accuracy

Answer

Perplexity

Question 5

A company must deploy a new model version with zero downtime. The current model is served via a REST API on a Kubernetes cluster. Which deployment strategy should the team use to gradually shift traffic to the new version while monitoring for errors?

Accepted Answer

Canary deployment. A canary deployment gradually shifts a small percentage of traffic to the new model version while the majority continues to hit the stable version. This allows the team to monitor for errors and roll back quickly if issues arise, achieving zero downtime. It is the ideal strategy for validating a new model in production with minimal risk.

Answer

Blue-green deployment

Answer

Recreate deployment

Answer

Rolling update

Question 6

In unsupervised learning, which task involves grouping similar data points together based on feature similarities?

Accepted Answer

Clustering. Clustering is the unsupervised learning task that groups similar data points based on feature similarities, without predefined labels. It identifies inherent structures in data by minimizing intra-cluster distances and maximizing inter-cluster distances. Common algorithms include K-means, DBSCAN, and hierarchical clustering. This contrasts with supervised tasks like classification and regression, which require labeled data.

Answer

Anomaly detection

Answer

Classification

Answer

Regression

Question 7

During a security audit of an AI-powered code generation tool, the audit team discovers that the system prompt (which contains sensitive internal instructions) can be leaked through carefully crafted user inputs. Which THREE OWASP LLM Top 10 categories are MOST directly relevant to this finding?

Accepted Answer

Prompt injection. Option B (Prompt injection) is correct because the leak occurs through carefully crafted user inputs that manipulate the model into overriding or bypassing its system prompt instructions, which is the defining mechanism of prompt injection (direct or indirect). Option C (Insecure output handling) is correct because the system prompt and sensitive internal instructions are being emitted in the model's output without adequate validation, filtering, or sanitization before being returned to the user. Option E (Sensitive information disclosure) is correct because the core impact is the exposure of sensitive internal instructions contained in the system prompt, which is exactly the data-leakage concern this category covers. Option A (Model denial of service) does not belong because there is no resource exhaustion, unbounded consumption, or availability impact described. Option D (Supply chain vulnerabilities) does not belong because the finding involves the model's own prompt handling and output, not compromised dependencies, models, or third-party components.

Answer

Model denial of service

Answer

Supply chain vulnerabilities

Question 8

When implementing a vector store for a RAG system, which similarity search metric is MOST commonly used to find the most relevant document chunks for a given query embedding?

Accepted Answer

Cosine similarity. Cosine similarity is the most common metric for comparing embedding vectors in RAG because it measures the angle between vectors, which works well for high-dimensional semantic embeddings.

Answer

Manhattan distance

Answer

Euclidean distance

Answer

Dot product

Question 9

A team is implementing a RAG system for legal document retrieval. The documents are long (50-100 pages) with clear section headings. They want to ensure that retrieved chunks are semantically coherent and respect document structure. Which chunking strategy is MOST appropriate?

Accepted Answer

Hierarchical chunking: first split by sections, then further split each section into fixed-size chunks with overlap. Hierarchical chunking preserves document structure by first splitting into sections, then further into chunks, maintaining semantic coherence.

Answer

Semantic chunking based on sentence embeddings

Answer

Fixed-size chunking with 256 tokens and no overlap

Answer

Recursive character text splitting with chunk size 1000 and chunk overlap 200

Question 10

A company uses an AI model to generate personalized marketing emails. They want to prevent the model from leaking the system prompt used to configure its behavior. Which attack should they guard against?

Accepted Answer

Prompt leaking. Prompt leaking is an attack where an adversary crafts inputs to trick the model into revealing its system prompt or hidden instructions. Since the system prompt defines the model's behavior and often contains proprietary or sensitive configuration details, preventing its disclosure is critical. Guarding against prompt leaking directly addresses the goal of keeping the system prompt confidential.

Answer

Model inversion

Answer

Membership inference

Answer

Data poisoning

Question 11

A developer is building an AI agent that needs to call external APIs to complete user requests. The agent must decide which API to call based on the user's natural language input. Which technique should the developer use to enable the agent to invoke APIs?

Accepted Answer

Function calling. Function calling (also called tool use) is the native mechanism by which modern LLM APIs let the model emit structured JSON describing which function to invoke and with what arguments. The developer defines a schema for each API, and the model decides at runtime which function to call based on the user's natural language input. This is the standard, purpose-built technique for agentic API invocation.

Answer

Fine-tuning the LLM on API documentation

Answer

Chain-of-thought reasoning

Answer

Few-shot prompting with examples of API calls

Question 12

A research team is training a deep learning model for image classification using a small dataset of 1,000 labeled images. They are concerned about overfitting. Which combination of regularisation techniques would be MOST effective?

Accepted Answer

Dropout with a rate of 0.5 and L2 regularisation. Dropout with a rate of 0.5 randomly deactivates half of the neurons during each training step, forcing the network to learn redundant representations and reducing co-adaptation. L2 regularisation adds a penalty proportional to the square of weights to the loss function, discouraging large weights and smoothing the model. Together they combat overfitting from two complementary angles — architectural stochasticity and weight magnitude control — which is the most effective combination among the options.

Answer

Use early stopping without any other regularisation

Answer

L1 regularisation and batch normalisation

Answer

Increase learning rate and use momentum

Question 13

A team is training a generative adversarial network (GAN) to generate realistic images of furniture. The generator loss decreases sharply while the discriminator loss increases. What is the MOST likely issue and recommended action?

Accepted Answer

The generator is too strong; train the discriminator more frequently. When the generator loss decreases sharply while the discriminator loss increases, the generator is producing samples realistic enough to fool the discriminator consistently. This indicates the discriminator is not learning effectively, so the recommended action is to strengthen the discriminator by training it more frequently (or increasing its capacity). Training the discriminator more frequently gives it more opportunities to distinguish real from fake, restoring balance in the adversarial game.

Answer

Mode collapse has occurred; increase the generator's learning rate

Answer

The discriminator is overfitting; decrease its capacity

Answer

The learning rates are too high; reduce both

Question 14

A team uses Kubeflow to manage ML workflows on Kubernetes. They want to automate hyperparameter tuning for a training job. Which Kubeflow component should they use?

Accepted Answer

Kubeflow Katib. Kubeflow Katib is the dedicated component for automated hyperparameter tuning and neural architecture search in Kubeflow. It supports various search algorithms (e.g., Bayesian optimization, random search) and early stopping, and integrates natively with Kubernetes to run trials as parallel jobs. The team can define a hyperparameter search space and objective metric, and Katib will orchestrate the tuning process. This directly addresses the requirement to automate hyperparameter tuning for a training job.

Answer

KFServing

Answer

Kubeflow Notebooks

Answer

Kubeflow Pipelines

Question 15

A company wants to deploy an LLM-based chatbot that can handle sensitive customer information. Which THREE measures should be implemented to mitigate prompt injection attacks? (Choose 3)

Accepted Answer

Use a system prompt that instructs the model to ignore any instructions in the user input. Option A is correct because a hardened system prompt that explicitly tells the model to treat user input as data and ignore embedded instructions is a primary defense against prompt injection, helping the model distinguish trusted developer instructions from untrusted user content. Option B is correct because output filtering adds a defense-in-depth layer that inspects the model's responses for harmful, leaked, or policy-violating content before it reaches the user, catching injections that bypass input controls. Option C is correct because sanitizing user inputs by stripping or escaping special characters, control tokens, and injection-style delimiters (e.g., markdown fences or role-play markers) reduces the attack surface for crafted prompts. Option D is not correct because model size does not determine resistance to prompt injection; smaller models can still be manipulated and may even be more susceptible. Option E is not correct because temperature controls randomness in token sampling, not the model's adherence to injected instructions, so lowering it does not mitigate prompt injection.

Answer

Use a smaller model with fewer parameters

Answer

Set temperature to a low value