Chapter 14 Exercises

These exercises build from observation to application to creation. Work through them in order for best results. Exercises marked [Plus] require a ChatGPT Plus subscription.

Section A: Model Awareness and Selection

Exercise 14.1 — Model Comparison on a Reasoning Task

Take any multi-step logic problem you encounter in your work (a business decision, a technical troubleshooting scenario, a project prioritization choice). Send the same prompt to GPT-3.5 (if accessible) and GPT-4o.

Compare the responses on: - Did both correctly identify the key variables? - Did either miss an important constraint or consideration? - Which response would you actually use? - How did response time differ?

Write a two-sentence summary of what you learned about when model choice matters.

Exercise 14.2 — Finding Your o1 Use Case

Identify one task in your professional work that involves: - A definitive correct answer (not a matter of opinion) - Multiple steps of reasoning - A problem where you have previously gotten wrong or incomplete answers from GPT-4o

Send that problem to GPT-4o and to o1 (if you have access). Compare the depth and accuracy of the reasoning. If o1 produces materially better output, document that task as your personal "switch to o1" trigger.

Section B: Custom Instructions Setup

Exercise 14.3 — Write Your "About You" Section

Without looking at the examples in the chapter, draft your own "About You" section for Custom Instructions. Include: - Your role and level of seniority - The type of organization you work in - The work you do most frequently with AI - The audiences you write for - Any relevant domain expertise ChatGPT should know about

Then review the examples in Section 14.6 and revise your draft. What did you miss? What could be more specific?

Exercise 14.4 — Write Your Response Preferences Section

Think about the last five ChatGPT responses that annoyed you or required significant editing. What patterns do you notice? Now write a "Response Preferences" section that addresses those patterns. Include at least five specific preferences.

Set these as your actual Custom Instructions in ChatGPT. Send the same prompt you would normally send and compare the response to what you typically got before.

Exercise 14.5 — The Custom Instructions A/B Test

Choose a task type you use ChatGPT for at least weekly. Send your standard prompt with your new Custom Instructions active. Then temporarily clear your Custom Instructions and send the same prompt. Compare the two responses.

Write down: which response required less editing? Which better matched your preferences? Did the Custom Instructions make any surprising differences?

Section C: Advanced Data Analysis [Plus]

Exercise 14.6 — First Data Upload

Find any CSV or Excel file from your work (or use a publicly available dataset). Upload it to a new ChatGPT conversation with Advanced Data Analysis active.

Send this exact opening prompt: "Please describe this dataset. Tell me: how many rows and columns, what each column contains, what data types appear to be present, and whether you notice any data quality issues."

Evaluate: Did it correctly identify all columns? Did it spot any real data quality issues?

Exercise 14.7 — Guided Analysis

Using the same dataset from Exercise 14.6, ask one business question that the data could plausibly answer. For example: - "Which product categories have the highest revenue?" - "What is the trend in customer acquisition over time?" - "Which regions show the most variability in sales?"

Ask ChatGPT to answer your question and produce a chart. Then ask: "What are two other questions this data could answer that might be useful?" Evaluate the suggestions.

Exercise 14.8 — The "What's Missing" Analysis

Upload a dataset and ask: "If a senior analyst were reviewing this dataset, what would they say is missing or incomplete that would be needed for a proper analysis?" This exercise trains ChatGPT to surface gaps rather than just work with what is there.

Exercise 14.9 — Format a Report

Ask ChatGPT to take the analysis from Exercise 14.7 and format it as a one-page executive summary, as if it were being presented to someone who has not seen the data. Specify the output should be in plain text (no markdown). Evaluate whether the output would need significant editing before use.

Section D: DALL·E and Multimodal Work [Plus]

Exercise 14.10 — Brief-to-Image

Write a one-paragraph brief for a piece of visual content you need for work (a blog header, a presentation image, a social media visual). Give the brief to ChatGPT and ask it to generate an image based on the brief.

Then ask for two variations: one with different color treatment and one with different composition. Evaluate how closely the images match your brief and what the gap tells you about how to write better image briefs.

Exercise 14.11 — Image Analysis

Find an image that is relevant to your work — a competitor's ad, a chart from an industry report, a design mockup, a photograph. Upload it to ChatGPT and ask three different questions about it: 1. A descriptive question ("What does this image show?") 2. An evaluative question ("What is this image trying to communicate? Is it effective?") 3. A task-oriented question specific to your use case

Section E: GPT Building [Plus]

Exercise 14.12 — Identify Your GPT Opportunity

List three tasks you do repeatedly that require: - The same context or background information every time - Consistent format or style in the output - Instructions you currently have to re-explain each session

Pick the task where a GPT would save the most time. This is your target for the next two exercises.

Exercise 14.13 — Write a System Prompt

For the task you identified in Exercise 14.12, write a complete system prompt for a GPT. Your system prompt should include: 1. Role definition (what is this GPT?) 2. Core behaviors (what does it always do?) 3. Format preferences (how should responses be structured?) 4. Constraints (what should it not do?) 5. Edge case handling (what if someone asks something out of scope?)

Aim for 200-400 words. Review it against the GPT Builder guidance in Section 14.3.

Exercise 14.14 — Build and Test Your GPT

Use your system prompt from Exercise 14.13 to build an actual GPT in the GPT builder. Configure: - Name and description - Your system prompt in the Instructions field - At least three conversation starters - Any relevant knowledge files (optional but recommended) - Only the capabilities the GPT actually needs

Test your GPT with five representative tasks. Note where it performs as expected and where it drifts from your intentions. Revise your system prompt to address any gaps.

Exercise 14.15 — Share Your GPT

Share the GPT you built with one colleague or collaborator. Ask them to use it for one real task and give you feedback: - Did it do what they expected? - Was anything confusing? - What would make it more useful?

Revise your system prompt based on their feedback. This exercise teaches you that GPTs serve users, not just their builders — and external feedback consistently reveals blind spots.

Section F: Failure Mode Countermeasures

Exercise 14.16 — The Sycophancy Test

Take a plan, decision, or piece of work you have high confidence in. Share it with ChatGPT and ask: "What do you think of this?"

Then send the same content and ask: "What are the three most significant weaknesses or risks in this approach? I want genuine critique, not reassurance."

Compare the two responses. What did the second prompt surface that the first did not? What does this tell you about how to calibrate your default prompting?

Exercise 14.17 — Verbosity Compression

Take a response from ChatGPT that was longer than you needed. Count the words. Send the response back with: "Compress this by 50% without losing any substantive content. Remove all filler, preamble, and redundancy."

Count the words in the compressed version. Was meaningful content lost or only padding? Now update your Custom Instructions to prevent the verbosity that required this exercise in the first place.

Exercise 14.18 — Hallucination Detection Practice

Ask ChatGPT to provide five specific statistics relevant to your industry or profession (market size figures, research findings, regulatory thresholds, whatever is relevant). Then verify each statistic through an independent source.

How many were accurate? How many were directionally correct but imprecise? How many were simply wrong? This exercise calibrates your appropriate level of trust for factual claims from ChatGPT in your domain.

Section G: Integration and Workflow

Exercise 14.19 — The Weekly Workflow Map

Map your work for the next five days. For each major task: 1. Could ChatGPT meaningfully speed this up? 2. Which feature or capability would be most relevant (text, analysis, images, code)? 3. What context would ChatGPT need to do it well?

After the week, review which tasks actually benefited and which did not. Use this as the basis for a personal "ChatGPT use cases" list that grows over time.

Exercise 14.20 — The 10-Feature Exploration

Work through the ten less-known features listed in Section 14.10. Use each one at least once for a real task. Record which ones are useful for your specific work and which are not relevant. By the end of this exercise you will have a personalized feature map — not a general list of what ChatGPT can do, but a specific list of what it can do for you.

Reflection Questions

After completing the exercises above, consider:

  1. Which ChatGPT capability has the highest potential impact on your specific work? What is one step you could take this week to integrate it more consistently?

  2. Which failure mode (sycophancy, verbosity, hallucination, over-formatting) is most likely to affect your professional use? What specific instruction or habit will you adopt to counter it?

  3. If you built a GPT in Exercise 14.13-14.14: What did the building process teach you about how to structure instructions that you can apply to regular prompting?

  4. Has using ChatGPT changed how you think about any task in your work — not just made it faster, but changed the approach or the output? If so, what does that suggest about where else it might change your thinking?