Chapter 3: Exercises

Below are five exercises intended to deepen your understanding of the core concepts from Chapter 3.

Each exercise is intended for the free version of Claude (by Anthropic), available at https://claude.ai, taking advantage of Artifacts to produce tables and comparisons in a visual format. The instructions will prompt you to refine your input and evaluate the AI’s outputs critically.

Exercise 1: Comparing Predictive vs. Reasoning Models

Purpose: To demonstrate the differences between a predictive AI model (e.g., GPT-4o) and a reasoning AI model (e.g., o1 or o3) by generating and refining outputs for different legal tasks.

Instructions:

  1. Initial Prompt to Claude
    Use a prompt asking Claude to produce a side-by-side comparison table (via Artifacts) of how a predictive model would handle drafting a simple legal disclaimer versus how a reasoning model would approach the same task. For example:
    Create a table comparing how a predictive AI model (like GPT-4o) and a reasoning AI model (like o3) would each handle drafting a short legal disclaimer for a website. Focus on speed, depth of analysis, and potential limitations. Format your response as an Artifact table.
  2. Refine the Prompt
    If the table is too shallow or lacks detail, refine your request:
    Please include specific rows on: - Average response time - Quality of legal reasoning - Style of language (formal vs. informal) - Risk of errors or omissions - Ideal use cases

Reflection:
Did Claude successfully produce a clear table with distinct comparisons between predictive and reasoning models? Note any areas where you might require more depth or specific details to fully understand each model’s strengths and weaknesses.

Exercise 2: Proprietary vs. Open-Source LLM Decision Matrix

Purpose: To evaluate proprietary (e.g., ChatGPT 4o) versus open-source LLMs in a hypothetical law firm setting and practice generating a structured visual comparison.

Instructions:

  1. Scenario Setup
    Imagine you work at a small firm with moderate IT support. The firm is debating whether to use a proprietary model like ChatGPT 4o or an open-source alternative. Consider factors like cost, ease of customization, vendor lock-in, transparency, data security, and support.

  2. Initial Prompt to Claude
    Using Artifacts, produce a decision matrix table comparing a proprietary model (e.g., ChatGPT 4o) and an open-source LLM in six categories: Cost, Customization, Data Security, Vendor Lock-In, Required Expertise, and Support.

  3. Refinement
    If the first table is too generic, add clarifying details:
    Please add a column for each category indicating whether it is "High," "Medium," or "Low" for each model. Then provide a short note under the table about key trade-offs.

Reflection:
Does the decision matrix capture your law firm’s main considerations? Which factors might change if the firm grows or if regulations become stricter?

Exercise 3: Identifying When to Use a Copilot vs. an Agent

Purpose: To clarify the differences between AI copilots (which assist humans but require oversight) and AI agents (which can operate more autonomously). You will create a table listing legal tasks suited to each approach, then refine the outputs.

Instructions:

  1. Scenario Prompt
    Create an Artifact table titled "When to Use a Copilot vs. an Agent in Legal Practice." Include at least five tasks for each category, and briefly explain why a copilot or agent is more suitable for each task.
  2. Refine the Prompt
    If the table lacks context or depth, refine it:
    Add two more columns: "Level of Autonomy Needed" and "Level of Ethical Risk," both rated on a scale of 1–5. Summarize how oversight differs between copilot tasks and agent tasks.

Reflection:
Look at the list of tasks under each category. Which tasks would you personally feel comfortable delegating to an agent? Did the table clarify the distinctions clearly enough for you to explain them to a non-technical colleague?

Exercise 4: Evaluating Real-World AI Tool Options

Purpose: To apply the material from Chapter 3 about ChatGPT 4o, Anthropic Sonnet 3.5 (Claude), Google Gemini, and Perplexity to a concrete scenario, generating a table in Claude’s Artifacts to compare them at a glance.

Instructions:

  1. Initial Prompt
    Assume you need an AI tool for a mid-sized law firm primarily focused on litigation support and basic contract reviews. Produce an Artifact table comparing ChatGPT 4o, Anthropic Sonnet 3.5, Google Gemini, and Perplexity across: Pricing, Context Window, Strengths in Litigation, Strengths in Contract Review, Main Drawbacks
  2. Refinement
    If the table is missing specifics, refine:
    Please add a short note underneath the table explaining why each tool’s context window might matter for large e-discovery tasks. Also, clarify how each tool handles or references external sources of case law.

Reflection:
Does the comparison table align with what you learned in Chapter 3? Which tool would you choose for your scenario, and why?

Exercise 5: Practical Considerations Check-List

Purpose: To practice creating a comprehensive but concise checklist (in a visual Artifact format) of practical considerations, like usability, integration, cost-effectiveness, and security, before choosing an AI tool.

Instructions:

  1. Scenario Prompt
    Generate an Artifact checklist titled “Key Practical Considerations for Adopting an AI Tool in a Law Firm.” Include at least five main considerations (usability, integration, cost, security, privacy) and provide 1–2 bullet points under each category summarizing Chapter 3’s key takeaways.
  2. Refinement
    If you need more detail regarding data security or privacy:
    Please expand the “Security and Privacy” section with specific examples relevant to attorney-client privilege and data protection laws.

Reflection:
Did the checklist capture all the big-ticket concerns from Chapter 3? Which item on the list do you think is the most challenging to address in a real law firm environment?

Additional Reflection (Optional)

After completing these exercises, write a short paragraph about your experience using Claude’s Artifacts feature to generate tables and comparisons. Consider how the visual format helped, or didn’t help, in clarifying differences between models, tools, and considerations. Reflect on how this might be useful in real-world legal settings where quick, clear decision-making is crucial.

By working through these exercises, you will reinforce your understanding of predictive vs. reasoning AI, proprietary vs. open-source solutions, and copilot vs. agent roles, ensuring you have both conceptual knowledge and practical skills to evaluate emerging AI tools in the legal field.