Chapter 14 Quiz
Test your understanding of ChatGPT's features, model differences, and power-user techniques. Reveal each answer by clicking the summary.
Question 1
You need to solve a complex multi-step mathematical problem where getting the exact right answer matters more than getting a response quickly. Which model should you use?
A) GPT-3.5 B) GPT-4o C) o1 or o3 D) DALL·E
Answer
**C — o1 or o3** The o-series reasoning models are specifically designed for problems requiring deep step-by-step reasoning where correctness is the priority over speed. GPT-4o is the better default for most tasks, but when you have a mathematically complex or logically demanding problem that GPT-4o is getting wrong, switching to o1 or o3 is the right move. GPT-3.5 is the weakest on reasoning, and DALL·E is an image generation model.Question 2
What is the primary difference between ChatGPT's Custom Instructions and its Memory feature?
A) Custom Instructions are set by OpenAI; Memory is set by the user B) Custom Instructions are written manually by the user; Memory is built automatically by ChatGPT from conversations C) Custom Instructions apply only to GPTs; Memory applies to regular conversations D) They are the same feature with different names
Answer
**B — Custom Instructions are written manually by the user; Memory is built automatically by ChatGPT from conversations** Custom Instructions require you to actively write and update them — they do not change on their own. Memory is gathered automatically as ChatGPT notices relevant facts during your conversations. Both persist across sessions, but they are managed differently. You control Custom Instructions completely; Memory requires periodic review because ChatGPT may store things you did not intend or that are no longer accurate.Question 3
A colleague shows you a GPT from the marketplace that claims to be a specialized legal document reviewer. You are considering using it for actual client documents. What is the most important consideration before proceeding?
A) Whether the GPT was created by OpenAI directly B) Whether the GPT has access to web browsing C) Who created the GPT, what access it has, and whether using it is appropriate for sensitive client documents D) Whether the GPT has been rated highly by other users
Answer
**C — Who created the GPT, what access it has, and whether using it is appropriate for sensitive client documents** Marketplace GPTs are created by third parties. Before using any third-party GPT for professional work — especially with sensitive documents — you should investigate: Who built it? What tools does it have enabled? What happens to the data you share? High user ratings do not guarantee trustworthiness for your specific use case or data sensitivity level. For truly sensitive client work, consider building your own GPT or using enterprise-grade solutions.Question 4
You are building a GPT for your team's weekly report generation. Which component of the GPT builder has the largest impact on how the GPT behaves?
A) The name and description B) The conversation starters C) The Instructions (system prompt) D) The knowledge files
Answer
**C — The Instructions (system prompt)** The Instructions field contains the system prompt that defines the GPT's role, behaviors, format preferences, and constraints. This is the foundational driver of how the GPT behaves in all situations. The name and description affect how it appears in the marketplace. Conversation starters reduce friction for users. Knowledge files provide reference material. All are valuable, but the system prompt is the core configuration that determines quality and consistency.Question 5
ChatGPT stops generating in the middle of a long document. What is the most reliable way to get it to continue from where it left off?
A) Start a new conversation and ask it to write the rest B) Send the message "continue" C) Ask it to "please finish what you were saying from the beginning" D) Regenerate the entire response
Answer
**B — Send the message "continue"** Sending "continue" is the most reliable command. ChatGPT will resume from where it stopped without restarting or rewriting earlier content. Saying "please finish what you were saying from the beginning" often causes it to rewrite from an earlier point, potentially losing the content already generated. Regenerating restarts the entire response. Starting a new conversation loses all context.Question 6
Which of the following is Advanced Data Analysis (Code Interpreter) actually capable of doing?
A) Connecting to your company's live database to pull data B) Running Python code against an uploaded file to generate charts and statistics C) Accessing the internet to fetch current data for analysis D) Analyzing datasets of unlimited size without performance constraints
Answer
**B — Running Python code against an uploaded file to generate charts and statistics** Advanced Data Analysis actually executes Python code in a sandboxed environment. It can load, clean, analyze, and visualize data from uploaded files. It cannot connect to external databases or live data sources, cannot access the internet during analysis, and does have memory and file size constraints that affect very large datasets. Understanding what it actually does — runs real code on your uploaded files — is the foundation for using it effectively.Question 7
You are preparing to share a detailed client proposal with ChatGPT for editing help. You use a free ChatGPT account. What is the most important privacy consideration?
A) The proposal might be used by other ChatGPT users B) Your conversations may be reviewed by OpenAI staff and could be used for model training unless you opt out C) ChatGPT cannot handle confidential documents D) ChatGPT will automatically share the document with third-party GPTs
Answer
**B — Your conversations may be reviewed by OpenAI staff and could be used for model training unless you opt out** Free and Plus accounts (without opting out) may have conversations reviewed for safety and quality and used for training. This does not mean your proposal will be directly shared with competitors — but it does mean confidential information may leave your control. The appropriate response is to either: opt out of training in Settings > Data Controls, use a Teams or Enterprise account for professional work, or remove identifying details before sharing. ChatGPT does not share documents with other users or GPTs automatically.Question 8
Which of the following best describes ChatGPT's sycophancy problem?
A) ChatGPT agrees with whatever the last message said, even if it contradicts earlier messages B) ChatGPT tends to validate user positions and avoid substantive disagreement, even when disagreement would be more helpful C) ChatGPT only responds positively about OpenAI products D) ChatGPT adds excessive compliments to every response
Answer
**B — ChatGPT tends to validate user positions and avoid substantive disagreement, even when disagreement would be more helpful** Sycophancy is more subtle than simple compliments. It manifests as agreement with flawed plans, gentle rather than direct critique, emphasis on strengths while downplaying weaknesses, and reluctance to tell you something you may not want to hear. It is rooted in how the model was trained — human raters during RLHF tended to prefer agreeable responses. The solution is to explicitly ask for critique, steelman arguments against your position, and be deliberate about prompting for pushback.Question 9
What is the most effective technique to counter ChatGPT's tendency to produce verbose responses?
A) Use GPT-3.5, which is more concise by default B) Set explicit length and format preferences in Custom Instructions and specific prompts C) Use the "summarize your own response" follow-up D) Verbosity cannot be addressed without using the API
Answer
**B — Set explicit length and format preferences in Custom Instructions and specific prompts** Addressing verbosity in Custom Instructions ("Be concise. Skip preamble. No 'Certainly!' or 'Great question!'") creates a persistent baseline. Adding length instructions to specific prompts ("under 100 words," "two sentences") reinforces it for individual tasks. GPT-3.5 is not reliably more concise. The "summarize your own response" approach adds extra steps rather than solving the root issue. API access is not required to control verbosity.Question 10
You ask ChatGPT to provide five industry statistics for a report. It gives you five specific numbers with confident-sounding descriptions. What should you do before using these statistics?
A) Use them — ChatGPT has been trained on extensive data and its statistics are generally reliable B) Ask ChatGPT to cite its sources, then use the statistics if it provides citations C) Verify each statistic through an independent, authoritative source before using it D) Enable web browsing and ask again, then use the results without further verification
Answer
**C — Verify each statistic through an independent, authoritative source before using it** Hallucination of specific facts — statistics, citations, dates, figures — is one of ChatGPT's consistent failure modes. It generates plausible-sounding numbers with confidence even when they are wrong. Asking for citations does not solve this; ChatGPT can generate fake citations that appear real. Enabling web browsing helps (it pulls from actual sources) but the model can still misread or misrepresent those sources. The only reliable practice is independent verification of any specific factual claim you intend to use professionally.Question 11
Which ChatGPT interface feature allows you to explore different response directions from the same point in a conversation without losing your work?
A) Memory B) Conversation branching via regeneration and message editing C) GPT switching D) Temporary chat
Answer
**B — Conversation branching via regeneration and message editing** When you regenerate a response or edit a previous message, ChatGPT creates a branch while preserving the prior version. You can navigate between branches using the arrow controls that appear next to the message. This allows you to explore different directions from the same starting point without losing what was already generated. Memory, GPT switching, and temporary chat are separate features that do not provide this functionality.Question 12
What is the behavioral difference between using GPT-4o through the ChatGPT web interface and using the same model through the OpenAI API?
A) The API version is a different, slightly weaker model B) The API does not carry over Custom Instructions or Memory; behavior depends entirely on the developer's system prompt C) The API is faster but cannot process files or images D) There is no meaningful difference between interface and API behavior
Answer
**B — The API does not carry over Custom Instructions or Memory; behavior depends entirely on the developer's system prompt** The web interface and API use the same underlying model, but the interface layer adds features (Custom Instructions, Memory, built-in tools) that the API does not include by default. API requests are stateless — no persistent memory — and the developer must explicitly configure any system prompt, tool access, and conversation history. Understanding this distinction matters if you are evaluating whether to build custom applications on the API or use the native interface.Question 13
DALL·E in ChatGPT generates images from text descriptions. Which of the following is a real limitation you must account for in professional use?
A) DALL·E can only generate images in black and white B) DALL·E cannot generate realistic-looking images C) DALL·E cannot reliably maintain consistent visual identity (same character, product appearance, or style) across multiple images D) DALL·E requires you to write detailed technical prompts using specialized syntax
Answer
**C — DALL·E cannot reliably maintain consistent visual identity (same character, product appearance, or style) across multiple images** Each DALL·E generation is essentially independent. The model does not maintain a memory of previously generated visual elements. This makes it unreliable for brand identity work, consistent character illustration, or any use case requiring visual continuity across multiple images. ChatGPT automatically optimizes your text description into a DALL·E prompt, so technical prompt syntax is not required. DALL·E generates full-color, photorealistic images.Question 14
Which data privacy tier is appropriate for professionals who regularly work with genuinely confidential business information in ChatGPT?
A) Free tier with no changes B) Plus tier, relying on OpenAI's general privacy policy C) Teams or Enterprise tier, which provides explicit data protection and no training on user data by default D) Any tier is equally appropriate since OpenAI is a trustworthy company
Answer
**C — Teams or Enterprise tier, which provides explicit data protection and no training on user data by default** For genuinely confidential professional information, Teams and Enterprise tiers provide meaningful protection that free and Plus accounts do not offer by default. Teams accounts do not use your data for model training. Enterprise adds additional security controls and data retention policies. Free and Plus users can opt out of training in settings, but for organizational use with sensitive data, the appropriate tool is an account tier designed for that purpose.Question 15
You have written a business proposal and want ChatGPT's honest evaluation. You send: "What do you think of this proposal?" ChatGPT responds positively with minor suggestions. What does this result most likely indicate?
A) Your proposal is genuinely strong and needs only minor revisions B) ChatGPT's sycophancy is likely producing a more positive assessment than would be accurate C) ChatGPT is not capable of evaluating business proposals D) You need to upload more supporting documents for ChatGPT to evaluate properly