Chapter 8 Quiz: Context Is Everything
Test your understanding of context loading, context types, and context management strategies. Attempt each answer before expanding the solution.
Question 1
Which of the following best describes the "blank slate problem" as introduced in this chapter?
A) AI models perform worse on complex tasks than simple ones B) Every new conversation session starts without memory of previous sessions, requiring context to be re-established C) AI models cannot retain information within a single long session D) AI models have no background knowledge about professional domains
Answer
**B) Every new conversation session starts without memory of previous sessions, requiring context to be re-established.** The blank slate problem is not about AI limitations in general (A) or within-session memory (C) — most models retain context within a single session. It is specifically about the absence of cross-session memory. Nor is it about background knowledge (D) — AI models have extensive world knowledge from training. The problem is that they have no knowledge of your specific situation, organization, or preferences unless you tell them, and they start fresh every session.Question 2
You receive AI output that is technically accurate but written at a vocabulary level far above your intended audience's comprehension. Which context type was most critically missing?
A) Background context B) Style context C) Audience context D) Constraint context
Answer
**C) Audience context.** Vocabulary level, depth of explanation, and assumed prior knowledge are all direct functions of audience context. Background context (A) provides situational information about you and your organization, not about the reader. Style context (B) addresses voice and tone, not reading level specifically. Constraint context (D) sets boundaries but does not proactively calibrate to a reader's knowledge level. The fix is explicit audience description: "Write for someone who has no technical background and reads at a general-audience level."Question 3
What is the primary purpose of the five-part context loading technique described in Section 8.3?
A) To reduce the length of individual prompts by front-loading all information B) To establish a structured, complete briefing before generating any output, preventing constraint violations and style drift C) To allow the AI to confirm its own instructions before responding D) To separate context from instructions so the AI can prioritize them separately
Answer
**B) To establish a structured, complete briefing before generating any output, preventing constraint violations and style drift.** The context loading technique is about front-loading completeness — ensuring all six context types are addressed before the first piece of output is generated, so that corrections happen before output rather than after. It reduces (not eliminates) prompt length by consolidating context once rather than repeating it per task (A). The optional confirmation step (C) is a useful extension but not the primary purpose. Separation of context and instruction (D) is a structural benefit, but not the primary purpose.Question 4
The "minimum effective context" principle states:
A) Always provide at least five sentences of context per prompt B) Context should be limited to the task description only C) Include only the context that would change the output in a meaningful way if removed D) Provide more context than you think is necessary, since more is always better
Answer
**C) Include only the context that would change the output in a meaningful way if removed.** The minimum effective context principle is a quality filter, not a quantity rule. There is no magic number of sentences (A). Restricting to task description only (B) ignores audience, style, and constraints. More is not always better (D) — excessive context dilutes focus, buries critical instructions, and can cause the AI to treat all information as equally relevant, reducing output quality.Question 5
According to the research discussed in Section 8.16, what is the "lost in the middle" effect?
A) AI models tend to forget earlier instructions when given new ones later in the conversation B) Information placed in the middle of a long context window receives less processing attention than information at the beginning or end C) AI models perform worse on middle sections of long documents than on introductions or conclusions D) Context provided mid-session is less reliable than context provided at the beginning
Answer
**B) Information placed in the middle of a long context window receives less processing attention than information at the beginning or end.** The "lost in the middle" effect is a documented characteristic of transformer-based models: they attend more reliably to content at the beginning and end of long input sequences than to content sandwiched in the middle. This has a direct practical implication: put your most critical instructions and constraints at the beginning of your prompt, not buried after paragraphs of background. Reference material and supporting context can follow the task statement.Question 6
What distinguishes a context packet from a one-time context load?
A) A context packet is shorter and simpler B) A context packet is a reusable, saved block of context designed to be pasted at the start of repeated sessions in the same domain C) A context packet includes more constraint information than a regular context load D) A context packet is used only at the system prompt level, not in conversational prompts
Answer
**B) A context packet is a reusable, saved block of context designed to be pasted at the start of repeated sessions in the same domain.** The defining characteristic of a context packet is reusability — you build it once and use it repeatedly, rather than reconstructing context from scratch at the start of every new session. It is not necessarily shorter (A), nor is it specifically constrained to constraint information (C) — it includes all relevant context types. It is used in conversational prompts, not only at system prompt level (D).Question 7
Which of the following best describes how to frame a reference document in a prompt?
A) Paste the document first, then ask your question at the end B) Introduce the document with what it is, how to use it, and which sections to prioritize, then paste it after your task description C) Ask the AI to summarize the document first, then use the summary as context for your actual task D) Always limit reference documents to 500 words to avoid overwhelming the model
Answer
**B) Introduce the document with what it is, how to use it, and which sections to prioritize, then paste it after your task description.** Framing transforms how the AI uses reference material. Without framing, the AI may summarize, respond to, or address the document broadly rather than using it as specified. The task statement should come first (not the document), and the framing should tell the AI the document's purpose and scope. Asking for a summary first (C) adds an extra round trip when direct framing accomplishes the same orientation. Word limits (D) may be appropriate in some cases but are not a general rule for reference material.Question 8
You are 40 exchanges into a long AI working session and notice the output has become increasingly formal, matching your original style guidance less and less. What is the most likely cause?
A) The AI model updated during your session B) The context window was exceeded and the model lost all earlier instructions C) Attention weight given to early session instructions can degrade over many exchanges, causing style drift D) The model has reached its response quality ceiling
Answer
**C) Attention weight given to early session instructions can degrade over many exchanges, causing style drift.** Style drift in long sessions is a well-documented phenomenon. It does not require the full context window to be exceeded (B) — it can happen well before that point as early instructions receive progressively less relative weight compared to more recent exchanges. Models do not update in real time (A), and there is no quality ceiling as described in (D). The solution is periodic context refresh — restating key style parameters to re-anchor the session.Question 9
True or False: The Custom Instructions feature in ChatGPT is a complete replacement for per-session context packets.
Answer
**False.** Custom Instructions are valuable for persistent, general-purpose context (your role, general preferences, default output formats). However, they are global — they apply to every conversation, which means project-specific or client-specific context cannot live there without potentially conflicting with other types of work. Task-specific, client-specific, and project-specific context still requires per-session loading or per-session packets. The two tools are complementary: Custom Instructions for consistent personal context, context packets for specific situational context.Question 10
What is the recommended structure for a prompt that includes both a task instruction and reference material?
A) Reference material → Task instruction B) Task instruction → Reference material with framing label C) Reference material → Summary request → Task instruction D) Task instruction → Summary request → Reference material
Answer
**B) Task instruction → Reference material with framing label.** The instruction-first rule is grounded in attention distribution research: the AI gives disproportionate weight to what appears first. Your task instruction should be the first thing the model reads, followed by clearly labeled reference material. Leading with reference material (A) can cause the AI to address the document broadly rather than using it to serve your specific task. The intermediate summary step in options C and D adds unnecessary rounds and is not a standard best practice for most reference-document tasks.Question 11
Which of the following style context approaches is most likely to produce output that closely matches your actual desired voice?
A) A list of five style adjectives: "professional, warm, direct, confident, accessible" B) A short description paired with a Do/Don't list and one sample sentence in the desired style C) A specification of reading level and sentence length D) A negative instruction: "do not write like a corporate communications department"
Answer
**B) A short description paired with a Do/Don't list and one sample sentence in the desired style.** The combination of description, Do/Don't pairs, and a concrete example gives the AI three different lenses through which to understand the target style. Adjective lists alone (A) are interpreted differently by different systems and rarely produce the specific effect intended. Reading level and sentence length (C) are useful dimensions but do not capture voice. Negative instructions (D) describe what you do not want without giving the AI a clear target to aim for.Question 12
Elena's client context packet (Section 8.13) includes a section called "Sensitivities (topics that require careful handling)." Which context type does this most closely represent?
A) Background context B) Style context C) Constraint context D) Audience context
Answer
**C) Constraint context.** Sensitivities are a form of situational constraint — things that exist in this specific client relationship that the AI would not know to navigate carefully without being told. They set boundaries on how the AI should handle certain topics, which is the defining characteristic of constraint context. Background context (A) provides situational information without specifying what to do differently. Style context (B) addresses voice and tone. Audience context (D) describes who the output is for, not what to avoid within the content.Question 13
What is the "context audit" technique used for?
A) Auditing an organization's AI usage policies B) Reviewing and updating a context packet when a project changes significantly C) Diagnosing disappointing AI output by identifying which context types were missing D) Checking whether a prompt's context exceeds the model's context window limit
Answer
**C) Diagnosing disappointing AI output by identifying which context types were missing.** The context audit is a post-output diagnostic tool. Rather than attributing poor output to AI limitations, the audit treats every disappointing output as an information gap to identify and close. It works through all six context types systematically, identifying which were absent and what specific information would have addressed each gap. It is not a policy audit (A), a document update process (B), or a technical token-counting tool (D).Question 14
Raj's codebase context packet includes a section called "Things We Have Consciously Decided." What prompting problem does this section solve?
A) It reduces the length of the context packet by consolidating multiple constraints B) It prevents the AI from flagging intentional patterns as bugs or recommending changes to established architectural decisions C) It tells the AI which parts of the codebase to prioritize in its review D) It provides examples of correct code for the AI to use as reference
Answer
**B) It prevents the AI from flagging intentional patterns as bugs or recommending changes to established architectural decisions.** Without this section, the AI would apply its general best-practice knowledge and flag items like "you should use asynchronous file reads" — which is correct in general, but wrong for Raj's codebase where synchronous reads are an intentional architectural choice. By explicitly labeling these decisions as intentional, Raj prevents the AI from wasting review time on false positives and prevents the AI from recommending changes that could break established patterns. This is a specific application of constraint context for technical review work.Question 15
Which failure pattern is described as: "Providing extensive context but an underspecified task, producing output that addresses the context broadly rather than accomplishing a specific goal"?
A) Assuming shared knowledge B) Context without task anchoring C) Forgetting to refresh in long sessions D) No style reference for style-sensitive work