Free PRACTICAL-APPLICATIONS-OF-PROMPT Practice Test Questions and Answers (2026)

The fundamental functionality of AI in processing raw data is to extract meaningful information and identify underlying structures. This enables core capabilities like pattern recognition, where the system learns to identify recurring arrangements in data, and classification, where it assigns data points to predefined categories. For example, an AI processes raw pixel data from an image to classify whether it contains a specific object. These capabilities are direct results of processing vast amounts of unstructured data.

[Cannot be determined without the full list of options.]

[Cannot be determined without the full list of options.]

[Cannot be determined without the full list of options.]

1. Russell, S. J., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson. (Chapter 1, Section 1.1, discusses AI capabilities; Chapter 19, discusses pattern recognition as a result of learning from data).

2. Stanford University. (n.d.). CS221: Artificial Intelligence: Principles and Techniques, Lecture Notes on Machine Learning. Retrieved from https://stanford-cs221.github.io/autumn2019/notes/ (Describes how models process raw data features for tasks like classification).

The scenario describes a "black-box" AI model, where the internal logic and decision-making criteria are not understandable to its users. This core issue is a lack of transparency, also known as a lack of explainability. When bank employees cannot understand why the AI suggested a particular outcome, they cannot verify its fairness, check for hidden biases, or be accountable for the final loan decision. This opacity is a significant ethical concern, particularly in high-stakes domains like finance where decisions profoundly impact individuals' lives.

A) Algorithmic bias is a potential consequence of the lack of transparency, but the fundamental problem described is the inability to inspect the model's reasoning.

C) Data privacy concerns the handling of applicant data, which is a separate issue from the model's operational transparency.

D) Job displacement is a broader socioeconomic impact of AI automation, not the specific ethical concern related to the model's inscrutable decision process.

1. Adadi, A., & Berrada, M. (2018). Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI). IEEE Access, 6, 52138-52160. Section II.A, "The Need for XAI," discusses how the lack of transparency in black-box models hinders trust and accountability. (https://doi.org/10.1109/ACCESS.2018.2870052)

2. Goodman, B., & Flaxman, S. (2017). European Union regulations on algorithmic decision-making and a "right to explanation". AI Magazine, 38(3), 50-57. This article directly links the "black box" problem to the need for explanations in automated decision-making systems. (https://doi.org/10.1609/aimag.v38i3.2741)

The most effective prompt engineering solution is to provide the model with specific, relevant context and a clear, direct instruction. To generate an accurate advertisement for a new sports watch, the prompt must include the factual features of that specific watch. This grounds the model's response in reality and prevents it from "hallucinating" or inventing features. Asking it to write an advertisement based on this provided information gives it a well-defined task. This approach combines providing context with a clear role and task, which is a best practice for achieving reliable and relevant outputs.

B. Ask the model to imagine a new sports watch: This will produce a generic advertisement with fabricated features, not one for the actual product.

C. Provide the model with advertisements for other products: This may help with style but fails to provide the necessary factual information about the new watch's features.

D. Ask the model for a list of common sports watch features: This is a preliminary research step and does not accomplish the final goal of creating the advertisement.

1. OpenAI. (2024). "Prompt engineering - Best practices for prompt engineering with OpenAI API." In the section "Provide reference text," the documentation states, "We can steer models to answer with respect to reference text... to help the model generate less fabricated answers."

2. Stanford University. (2024). "CS324 - Large Language Models, Prompting." The course material highlights the importance of "Grounding" by providing the model with relevant information to reduce factual inaccuracies. It also emphasizes giving the model a clear "Task."

Descriptive language is crucial for guiding generative AI models to produce images that align with the user's specific intent. By providing rich details about subjects, attributes, setting, composition, lighting, and artistic style, the user reduces ambiguity. This specificity constrains the model's vast possibility space, leading to a more accurate, coherent, and contextually appropriate visual output. Vague prompts result in generic or unpredictable images, whereas descriptive prompts enable fine-grained control over the final creation, effectively translating a mental concept into a precise digital image.

Vague Language: Using non-descriptive or generic terms fails to provide the AI with sufficient direction, leading to outputs that do not match the user's vision.

Overly Complex Syntax: While descriptive, using convoluted sentence structures can confuse the model; clarity and specificity are more important than grammatical complexity.

Irrelevant Information: Including details unrelated to the desired visual outcome can introduce unwanted artifacts or misinterpretations in the generated image.

1. Brown, T. B., et al. (2020). Language Models are Few-Shot Learners. Section 3.3 "In-context learning". The paper demonstrates how specific, descriptive examples (prompts) guide model behavior. (Available via arXiv:2005.14165)

2. OpenAI. (2023). OpenAI Cookbook: Techniques to improve reliability. Section "Prompting best practices". Official documentation emphasizes providing detailed and specific context in prompts for better results.

3. Stanford University. (CS224N). Lecture Notes on Natural Language Generation. These materials discuss how conditioning text (the prompt) directly controls the output of generative models, with more detailed conditions leading to more specific outputs.

The creation and circulation of deepfake videos showing public figures making false statements is a direct example of generating and propagating misinformation and disinformation. The primary ethical failure is the deliberate deception of the public by presenting fabricated content as authentic. This undermines trust in individuals, institutions, and the media, which is a core concern in the field of AI ethics. The technology is used specifically to create a believable lie, which directly addresses the challenge of maintaining an informed public discourse and preventing malicious manipulation.

Bias and Fairness: This concerns AI systems that produce systematically prejudiced outcomes against certain groups. The issue described is deception, not necessarily biased treatment.

Privacy: This would be the primary concern if the deepfakes were created using non-public, private footage without consent. Here, the main issue is the false narrative.

Accountability: While determining who is accountable for the deepfake is an important related issue, the fundamental ethical problem demonstrated by the content itself is misinformation.

1. Chesney, R., & Citron, D. (2019). Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security. Lawfare Institute, Paper No. 19-01, Section II.A, "Disinformation and Defamation." (Discusses deepfakes as a powerful tool for disinformation).

2. Stanford University Human-Centered Artificial Intelligence (HAI). (2023). AI Index Report 2023. Chapter 5: Public Opinion, "Misuse of AI." (Highlights public concern over AI's potential for spreading false information).

3. Floridi, L. (2018). The Cambridge Handbook of Information and Computer Ethics. Cambridge University Press. Chapter 12: "Information, Misinformation, and Disinformation." (Provides a foundational ethical framework for distinguishing and analyzing the harm of false information).

Generative AI's value in data classification stems from its ability to perform zero-shot learning. Pre-trained on vast corpora of text and data, these models develop a deep understanding of language and concepts. This allows them to classify data into categories they have never explicitly been trained on by evaluating the semantic similarity between the input data and the class labels. This drastically reduces the need for large, manually labeled training datasets.

A. It can only classify numerical data types: This is incorrect; generative AI excels at classifying unstructured data like text and images.

C. It replaces the need for human oversight in the classification process entirely: This is false. Human validation is essential to ensure accuracy, mitigate bias, and handle edge cases.

D. It reduces the computational cost of training models to zero: This is misleading. While it reduces labeling costs, inference with large models is computationally expensive.

1. Radford, A., et al. (2021). Learning Transferable Visual Models From Natural Language Supervision. Proceedings of the 38th International Conference on Machine Learning, PMLR 139:8748-8763. (This paper on CLIP is foundational to modern zero-shot classification).

2. Brown, T. B., et al. (2020). Language Models are Few-Shot Learners. Advances in Neural Information Processing Systems 33 (NeurIPS 2020). (The GPT-3 paper, which demonstrates few-shot and zero-shot capabilities, Section 3.8).

3. Hugging Face. (n.d.). Zero-shot classification. Hugging Face Documentation. Retrieved from the official Hugging Face documentation. (This vendor documentation provides a practical definition and implementation of the concept).

Transparency is the ethical principle requiring that the operations of an AI system be understandable to its users and stakeholders. It involves disclosing how the system works, the data it uses, and the logic behind its decisions. A key component of transparency is explainability, which is the ability to provide a human-understandable rationale for a specific output. This builds trust, allows for debugging, enables users to contest decisions, and helps ensure the system is operating fairly and as intended.

[Cannot be evaluated without the options provided].

1. Adadi, A., & Berrada, M. (2018). Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI). IEEE Access, 6, 52138-52160. https://doi.org/10.1109/ACCESS.2018.2870052 (See Section I, "Introduction," for a definition of XAI).

2. MIT OpenCourseWare. (2022). 6.S897: Machine Learning for Healthcare - Lecture 18: Fairness, Accountability, Transparency. Massachusetts Institute of Technology. (Defines and discusses the importance of transparency in high-stakes domains).

3. Doshi-Velez, F., & Kim, B. (2017). Towards A Rigorous Science of Interpretable Machine Learning. arXiv:1702.08608. (Provides a foundational academic perspective on the need for explainability).

Systematic bias, also known as measurement bias, refers to a consistent, repeatable error in a system's output. The AI model's prediction of temperatures being consistently "off by about five degrees" indicates a predictable flaw rather than a random error. This type of bias suggests a fundamental issue in the data collection, feature engineering, or the model's algorithm that causes its predictions to be skewed in the same direction and by a similar magnitude every time.

Confirmation bias: This is a user's tendency to favor information that confirms their beliefs, not a consistent error generated by the model itself.

Automation bias: This describes a human's over-reliance on an automated system's output, which is unrelated to the model's inherent prediction error.

Selection bias: This results from non-random data sampling during training. While it can lead to systematic errors, "systematic bias" is the more direct and precise term for the observed phenomenon.

1. Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A Survey on Bias and Fairness in Machine Learning. ACM Computing Surveys, 54(6), 1-35. https://doi.org/10.1145/3457607 (Section 2.2.2 discusses Measurement Bias, where data is skewed in a particular direction, leading to systematic errors).

2. Baeza-Yates, R. (2018). Data and Algorithmic Bias. In Encyclopedia of Big Data Technologies. Springer. https://doi.org/10.1007/978-3-319-63962-833-1 (Explains how systematic errors in data lead to biased models that consistently make incorrect predictions for certain groups or conditions).

3. MIT OpenCourseWare. (2022). 6.S191: Introduction to Deep Learning. Lecture 10: Fairness, Privacy, and Ethics. (The course discusses how systematic flaws in data collection lead to biased model performance).

Generative models rely on probabilistic token sampling, so the same prompt can yield different, sometimes fabricated, records. This non-deterministic “hallucination” makes it hard to guarantee that every element has been captured and sorted into the correct place, creating accuracy and consistency problems in data-sorting workflows.

A. Data privacy is a concern with any AI system, not a challenge unique to using generative AI for sorting.

C. Reduced processing time is a benefit, not a challenge.

D. Lower storage cost is an operational benefit, not a risk introduced by generative output.

1. Holtzman, A. et al., “The Curious Case of Neural Text Degeneration,” ICLR, 2020, §3.2.

2. Mitchel, M. et al., “Model Cards for Model Reporting,” FAT 2019, p.6–7.

3. EU High-Level Expert Group on AI, “Ethics Guidelines for Trustworthy AI,” 2019, r.2.1.

Prompt chaining is a technique where a complex task is broken down into a sequence of simpler, interconnected prompts. The output from an initial prompt is used as input or context to guide the large language model (LLM) in a subsequent prompt. This method allows for more controlled and accurate generation for multi-step problems by creating a workflow of prompts. Each step in the chain builds upon the previous one, leading to a more comprehensive final result than a single, complex prompt might achieve.

B. Iterative Prompting: This involves refining a single prompt multiple times to improve its output, not using one prompt's output to inform a second, different prompt.

C. Few-shot Prompting: This technique provides examples within a single prompt to guide the model's response format, not linking multiple prompts together.

D. Self-consistency: This involves running the same prompt multiple times and taking the most frequent answer to improve reasoning, not creating a sequence of prompts.

1. Wu, T., et al. (2023). "PromptChainer: Chaining Large Language Model Prompts through Visual Programming." arXiv preprint arXiv:2303.06566. Section 2.1, "Chaining Prompts," describes how the output of one LLM call can be used as the input for another. (https://doi.org/10.48550/arXiv.2303.06566)

2. Wang, L., et al. (2023). "LLM-powered Autonomous Agents." arXiv preprint arXiv:2309.07864. Section 3.1.1, "Task Decomposition," discusses breaking down tasks into smaller subgoals, which is the foundational principle of prompt chaining. (https://doi.org/10.48550/arXiv.2309.07864)