Module 1 Assessment Quiz | CPE

Q1 Part 1: Understanding LLMs

What is the fundamental way Large Language Models (LLMs) generate text?

A.Following pre-programmed rules and grammar structures B.Predicting the next most probable token based on patterns learned from training data C.Searching a database of pre-written responses D.Understanding language meaning like humans do

Explanation

LLMs are fundamentally statistical prediction machines. They predict the next token (word or subword) based on probability distributions learned from billions of examples in their training data. They don't truly "understand" language or follow explicit rules.

Q2 Part 1: Understanding LLMs

Which of the following is the MOST critical limitation of LLMs for legal professionals to understand?

A.They cannot process images B.They are expensive to use C.They respond slowly D.They can hallucinate plausible-sounding but completely fabricated information, including fake case citations

Explanation

Hallucination is the most dangerous limitation for legal work. LLMs can generate convincing but entirely fictitious case names, statutes, and legal principles. Multiple lawyers have faced sanctions for submitting AI-generated briefs with fake citations. ALWAYS verify independently.

Q3 Part 2: Prompt Anatomy

In the CRISP framework for prompt construction, what does the "S" stand for?

A.Summary B.Style C.Specifications (format, length, structure) D.Source

Explanation

CRISP stands for Context, Role, Instructions, Specifications, and Parameters. Specifications define the output requirements: format (bullet points, prose, table), length, structure, and style.

Q4 Part 2: Prompt Anatomy

Scenario

You ask an AI: "Help me with this contract." The response is generic and unhelpful.

What is the PRIMARY problem with this prompt?

A.Vague instructions -- doesn't specify what action to take or what aspect to focus on B.Too short C.Wrong model selected D.Missing punctuation

Explanation

The prompt suffers from vague instructions. "Help me" doesn't specify what action to take (review? draft? explain?), what aspect to focus on, or what you want to learn. The AI must guess your intent, leading to generic responses.

Q5 Part 3: Prompt Types

When should you use few-shot prompting instead of zero-shot?

A.When you want faster responses B.When you need a specific output format or consistent classification pattern C.When the task is very simple D.When you want to save tokens

Explanation

Few-shot prompting is ideal when you need the AI to follow a specific format or pattern consistently. By providing examples, you "teach" the model your requirements through demonstration, dramatically improving format adherence.

Q6 Part 3: Prompt Types

Which phrase is most effective for activating Chain-of-Thought reasoning?

A."Give me the answer quickly" B."Just tell me what to do" C."Let's think through this step by step" D."Be brief"

Explanation

"Let's think through this step by step" is a classic Chain-of-Thought trigger phrase. Research shows this simple addition can improve accuracy on complex reasoning tasks by 10-40% by forcing the model to show intermediate reasoning steps.

Q7 Part 4: Token Economics

Approximately how many tokens does 1,000 words of English text represent?

A.500 tokens B.1,000 tokens C.2,000 tokens D.1,333 tokens (approximately)

Explanation

The rule of thumb is 1 token ≈ 0.75 words. Therefore, 1,000 words ≈ 1,333 tokens (1000/0.75). Conversely, 1,000 tokens ≈ 750 words. This helps estimate context usage and costs.

Q8 Part 4: Token Economics

In API pricing, which typically costs MORE per token?

A.Output tokens (AI's response) B.Input tokens (your prompt) C.They cost the same D.System tokens

Explanation

Output tokens typically cost 2-5x more than input tokens. This is because generating text requires more computation than processing input. This is why requesting concise responses can significantly reduce costs.

Q9 Part 5: Common Pitfalls

Scenario

An AI provides a response with 5 case citations supporting a legal argument. What should you do FIRST?

What is the correct approach?

A.Use the citations immediately since AI is reliable B.Independently verify each citation exists and says what the AI claims C.Ask the AI to confirm the citations are real D.Check if the citation format looks correct

Explanation

You must independently verify each citation. LLMs frequently hallucinate case names, and asking the AI to confirm its own citations is unreliable. Several lawyers have faced sanctions for fake AI-generated citations. Always check in primary legal databases.

Q10 Part 5: Common Pitfalls

In the TEAR iterative refinement cycle, what does the "A" stand for?

A.Analyze B.Accept C.Adjust (make ONE focused change) D.Attempt

Explanation

TEAR stands for Test, Evaluate, Adjust, Repeat. The key is to make ONE focused change to address the most critical issue. Changing multiple things at once makes it impossible to know what worked.

Module 1: Foundations of Prompt Engineering

Instructions