Prompt Engineering and Human-Computer Interaction

Comprehensive practice flashcards covering Human-Computer Interaction evolution, prompt engineering principles, advanced prompting techniques (CoT, ToT, L2M), evaluation frameworks, and enterprise deployment strategies.

What is Human-Computer Interaction (HCI), and by what other names is it known?

HCI is a field of study that refers to communication between the human user and a computer system, concerned with design, evaluation, and implementation to provide a user-friendly environment. It is also known as Computer Human Interface (CHI) or Man Machine Interaction (MMI).

Which three core disciplines contribute to Human-Computer Interaction (HCI) as depicted in this diagram?

The three contributing disciplines are Computer Science, Cognitive Psychology (noted in the diagram as Cognitive Psycology), and Fine Arts Design.

What key historical HCI milestones occurred in the years $1984$, the $1990\text{s}$, and $2006$?

In $1984$, the first Macintosh was released with a revolutionary graphical user interface and multiple fonts. In the $1990\text{s}$, the Internet expanded and the World Wide Web (WWW) was created by Tim Berners-Lee. In $2006$, NINTENDO released the Wii featuring a rear remote controller that detects movements in 3D.

What are the five key characteristics of an effective prompt?

1. Clarity (simple, clear, unambiguous language)\n2. Specificity (explicit details and requirements)\n3. Context (background information, purpose, and target audience)\n4. Structure (systematic and logical organization)\n5. Expected Output (specified format, length, tone, or style)

What are the four core structural components of a comprehensive prompt?

1. Instruction (defines the main task or action with clear action words)\n2. Context (provides background information, target audience, or purpose)\n3. Input Data (the material or information to be processed)\n4. Output Indicator (specifies presentation details such as format, length, or headings)

How do Zero-Shot Prompting and Few-Shot Prompting compare in terms of examples provided and token cost?

Zero-Shot Prompting provides no examples, relies entirely on pre-trained knowledge, and has lower token usage and cost per request. Few-Shot Prompting provides a small set of input-output examples ($2\text{–}10$), providing higher accuracy and formatting consistency at the expense of higher token usage.

What is the function of the "Delimiters" technique in prompt engineering, and what are common examples?

Delimiters are special markers, symbols, or tags that separate instructions from context, input data, or examples. Common types include triple quotes (""" """), backticks (), XML-style tags ( ), brackets ([ ]), and custom markers (<<>>).

What are the main differences between Instruction-based prompting and Role-based prompting?

Instruction-based prompting focuses on the "what" using direct operational commands/steps for structured tasks. Role-based prompting focuses on the "who" by assigning a persona or identity that primes an expert knowledge framework and tone for perspective-driven outputs.

What is constraint-based prompting and what types of boundaries does it establish?

Constraint-based prompting adds explicit boundaries to control how the AI responds. Common constraints include word count, tone (e.g., formal, friendly), format (e.g., tables, bullet points, JSON), language complexity, and explicit exclusions or inclusions.

How does Chain of Thought (CoT) prompting differ from Least-to-Most (L2M) prompting?

Chain of Thought (CoT) encourages the model to generate a linear sequence of intermediate reasoning steps before reaching the final answer. Least-to-Most (L2M) prompting decomposes a problem into smaller sub-problems, solves them sequentially from easiest to hardest, and passes each intermediate result to the next sub-problem.

What are the steps of the Prompt Refinement Workflow shown in this diagram?

1. Identify the objective\n2. Find ambiguity\n3. Add context\n4. Make the task specific\n5. Specify the output format\n6. Add constraints\n7. Test the prompt\nDecision: "Output satisfies the goal?" (If No, loop back to Revise and improve; if Yes, reach 8. Final refined prompt).

How can prompt engineers identify and mitigate ambiguity and cognitive bias in prompts?

Ambiguity is mitigated by replacing vague terms with multiple meanings (e.g., "best", "fast") with explicit measurable criteria and context. Cognitive bias is mitigated by avoiding leading questions, removing unsupported assumptions, using neutral language, and asking for multiple perspectives and evidence.

What are the seven stages of the Error Identification and Correction Cycle in prompt engineering?

1. Initial Prompt (write prompt)\n2. AI Response (generate and observe)\n3. Identify Errors (check for incorrect/incomplete content)\n4. Refine the Prompt (modify by adding clarity/context)\n5. Test Again (run corrected prompt)\n6. Evaluate (check accuracy/relevance)\n7. Repeat if Needed

According to the Iterative Prompt Improvement Method, what are the typical iteration counts across different task complexities?

Simple tasks (e.g., summarization) typically require $1\text{–}2$ iterations; moderately complex tasks (e.g., specific format or tone) require $3\text{–}5$ iterations; and complex production tasks (e.g., multiple requirements, high reliability) require $5\text{–}10+$ iterations.

What is the Self-Criticism prompting technique and what is its primary limitation?

Self-Criticism instructs the AI to generate an initial response, evaluate its own output for errors, omissions, or instruction violations, and produce an improved final answer. Its primary limitation is that the same model evaluating itself may fail to detect its own factual errors or produce a flawed critique.

How do lower and higher settings for Temperature and Top-P alter AI outputs, and what tasks are they best suited for?

Lower values produce focused, predictable responses with narrower token selection, best suited for facts, coding, and structured tasks. Higher values produce creative, diverse responses with wider token selection, best suited for brainstorming and creative writing.

What are the six key steps for structuring a prompt to generate effective and unbiased surveys?

1. Define the Objective\n2. Specify the Target Audience\n3. List the Key Topics\n4. Choose Question Types\n5. Ensure Neutral and Clear Language\n6. Set Additional Details (number of questions, tone, length, instructions)

What are the seven steps outlined in the Problem-Solving and Decision-Making "Pros and Cons" Framework?

1. Identify the Problem\n2. List Options\n3. Find Pros\n4. Find Cons\n5. Set Criteria\n6. Compare Options\n7. Analyze & Decide (leading to a balanced and informed final decision)

What are the three primary ethical considerations when designing and deploying prompts?

1. Privacy: Avoiding confidential or sensitive data exposure by anonymizing inputs.\n2. Copyright: Respecting creators' intellectual property by asking for summaries/analyses rather than full unauthorized reproduction.\n3. Harmful Content: Preventing the generation of dangerous, illegal, or harassing material, framing sensitive inquiries around safety and awareness.

How do prompts maintain safety and accuracy in the medical and healthcare domain?

Healthcare prompts enforce strict boundaries: restricting the AI to use only verified provided records, forbidding data fabrication, requiring explicit declarations when information is missing, protecting patient privacy, prohibiting independent diagnosis or prescription, and requiring oversight from qualified clinicians.

What are domain-specific hallucinations and how does grounded prompting prevent them?

Domain-specific hallucinations occur when an AI fabricates false facts, citations, or legal/medical codes due to limited specialized knowledge. Grounded prompting prevents them by restricting the model strictly to provided source materials, demanding evidence/citations, and explicitly instructing the model to state uncertainty rather than guessing.

What do the letters in the CO-STAR prompt framework represent?

C – Context (background information)\nO – Objective (the core task to accomplish)\nS – Style (writing style)\nT – Tone (emotional attitude)\nA – Audience (intended reader/user)\nR – Response (expected output format)

What do the letters in the CREATE prompt framework represent?

C – Character (role or persona)\nR – Request (task to perform)\nE – Examples (demonstrations of desired output)\nA – Adjustments (constraints or modifications)\nT – Type (output format or medium)\nE – Extras (additional details or requirements)

How does an Agentic Workflow differ from a Single Prompt across key dimensions?

A Single Prompt consists of one instruction and response, has direct user specification, limited external tool interaction, single generation output, and low autonomy. An Agentic Workflow consists of multiple actions, goal-driven tool use (search, databases, APIs), iterative refinement, and autonomous planning.

What is a prompt template and why is it beneficial in software development?

A prompt template is a predefined prompt structure containing static instructions and dynamic placeholders. It ensures consistency across engineering teams, saves development time, enables standardized workflows (debugging, code generation, testing), and allows programmatic integration and automated CI/CD testing.

How is a prompt reverse-engineered from an existing AI-generated output?

Reverse engineering involves examining the output to deduce the likely objective, background context, voice, tone, target audience, constraints (such as word count or language), and format, followed by reconstructing and testing a candidate prompt against the model.

What is the Tree of Thoughts (ToT) framework and how does it advance beyond Chain of Thought (CoT)?

Tree of Thoughts (ToT) extends Chain of Thought by allowing LLMs to explore multiple alternative reasoning paths simultaneously in a tree structure. It generates candidate intermediate thoughts, evaluates their feasibility, searches branches using strategies like BFS or DFS, and backtracks when a branch fails.

How do System Prompts differ from User Prompts in LLM interactions?

A System Prompt establishes the overall persona, behavioral guidelines, constraints, tone, and global interaction context ("How should the AI behave?"). A User Prompt specifies the immediate task, question, or request for the model to execute ("What should the AI do?").

What role do Prompt Version Control and Prompt Management Systems serve in enterprise deployments?

Prompt Version Control tracks changes, enables side-by-side comparison, and permits rollback of prompt revisions like source code. Prompt Management Systems provide centralized libraries for storing, testing against objective benchmarks (e.g., accuracy, latency), monitoring costs, and safely deploying prompts across production pipelines.

What is the primary pedagogical principle of the personalized Socratic tutor prompt framework?

The Socratic tutor framework instructs the AI to guide learners to discover solutions through progressive questioning, asking one question at a time and offering incremental hints rather than immediately providing direct answers.