AI Basics Quiz: What Do You Actually Know About AI?

What is the 'basic prompt improvement loop'?

Copy your prompt from online prompt libraries rather than writing your own Send a prompt, notice what's missing or wrong in the response, add that context and re-prompt, repeat — treating AI interaction as iterative rather than single-shot Always delete and restart a new conversation if the first response isn't perfect Submit the same prompt three times and choose the best of the three responses

Which of the following is an example of AI that predates ChatGPT and has been running invisibly for years?

ChatGPT Plus, which was the first commercial AI product available to consumers A basic spreadsheet formula that calculates the sum of a column A credit card fraud detection system that flags unusual transaction patterns — a classic machine learning application deployed by banks for over a decade A dropdown menu that shows you your most recently visited websites

What is AI 'hallucination'?

When AI becomes confused and stops generating responses When AI translates text incorrectly because it doesn't know the language When AI generates plausible-sounding but factually incorrect content — stated confidently, without any signal to the user that the information may be wrong When AI produces visual artifacts or distorted images

What happens when a conversation exceeds an AI's context window limit?

The AI stores the overflow in external memory and retrieves it when needed The AI loses access to text that falls outside the window — it effectively 'forgets' earlier parts of the conversation and may give inconsistent responses Exceeding the context window triggers an automatic upgrade to a larger model The AI stops responding and requires a new conversation to be started

According to McKinsey's 2025 survey, approximately what fraction of organizations report regular AI use in at least one business function?

88% — AI is now mainstream in organizations, not experimental Less than 10% — AI is still primarily used only by technology companies About 25% — AI adoption is still in early phases for most organizations About 50% — half of organizations have started; half remain uninvolved

For which type of task would AI hallucination risk be LOWEST?

Describing a competitor's product features from their latest press release Looking up the exact revenue of a private company last quarter Brainstorming a list of marketing tagline ideas — there are no 'correct' facts to get wrong; creativity and fluency are the objective Finding the specific statute number of a regulation passed in 2025

Which of the following tasks is AI MOST reliable at?

Telling you whether a news article published yesterday is accurate — AI can't verify real-world current facts Calculating the exact compound interest on a complex loan schedule — mathematical precision is a known AI weakness Producing a well-structured first draft of a professional email — AI draws on vast writing patterns and excels at fluent, structured text generation Accurately citing a specific legal case from last month — AI has knowledge cutoffs and hallucination risk with specific facts

Which of the following is the correct relationship between AI, machine learning, and generative AI?

Generative AI is a type of machine learning, which is a type of AI — they are nested categories, not separate things AI is only about robots; machine learning is about software; generative AI is a marketing term Machine learning and generative AI are the same thing with different names Generative AI came before machine learning, which came before AI

What does the Gartner Hype Cycle predict about technologies that go through a 'trough of disillusionment'?

The trough of disillusionment only applies to hardware technologies, not software like AI Technologies that enter the trough of disillusionment never recover and should be abandoned The Hype Cycle predicts AI will be replaced by a completely different technology by 2027 Most eventually reach a 'plateau of productivity' where real, consistent value is delivered — the trough represents reality calibration, not permanent failure

What is the transformer architecture, and why does it matter for modern AI?

A type of electrical transformer that powers AI data centers more efficiently A 2017 research breakthrough that enabled models to track how words relate across long text — making it possible to train AI on vast amounts of text at a scale that produces today's LLMs A programming language used to build ChatGPT and similar AI products A hardware chip designed specifically for Google's AI research

What is the key difference between 'narrow AI' and 'general-purpose AI'?

Narrow AI runs on phones; general-purpose AI requires large servers Narrow AI is for small companies; general-purpose AI is only for enterprise use Narrow AI is designed for one specific task (spam filtering, recommendations); general-purpose AI handles many different tasks through a single model that users direct with language Narrow AI is cheaper; general-purpose AI is always more expensive per use

Which of the following describes a realistic and high-value use of AI in 2026?

Using AI to replace all written communication with a client without any human involvement Delegating an entire client project to an AI agent with no human review of outputs Using AI to produce a first draft of a client proposal, then editing it with domain knowledge and relationship context the AI doesn't have Using AI to make final legal or medical decisions without a professional reviewing them

What is a 'token' in the context of AI language models?

A single character of text — AI processes text one character at a time A unit of text roughly corresponding to a word or word fragment — the basic unit AI models use to process and generate language, and how usage and pricing are measured A cryptocurrency used to pay for AI model usage A security credential required to access AI APIs

Which of these everyday products uses machine learning or AI?

Netflix's recommendation engine — it uses machine learning to predict what content you'll enjoy based on your viewing patterns A basic alarm clock — it does only what it's explicitly programmed to do A standard calculator — it follows fixed mathematical rules with no learning A physical thermostat with a timer — it executes pre-set rules without learning

What made ChatGPT feel fundamentally different from previous AI products like Siri or Alexa?

ChatGPT was the first AI to use the internet in real time to answer questions ChatGPT was the first AI ever to understand English language ChatGPT was a general-purpose system — one interface could write, analyze, code, and answer questions across any domain, unlike assistants built for narrow command-response tasks ChatGPT was different because it ran on quantum computers unavailable to earlier AI

When would you want a LOWER temperature setting for an AI task vs. a HIGHER temperature?

Temperature only affects response speed, not content — lower is faster, higher is slower Temperature controls language formality: low for formal writing, high for casual writing Lower for precise tasks (data extraction, code, factual answers) where consistency matters; higher for creative tasks (brainstorming, varied options, creative writing) where variation is desirable Higher temperature always produces better quality — you should maximize it for all tasks

Which of the three habits of effective AI users is most important for working with AI in a professional context?

Using AI for every task, no matter how simple, to build maximum fluency Delegating final decisions to AI to remove human bias from the process Keeping the judgment — AI generates options and drafts; the human remains accountable for decisions and final outputs Never verifying AI output to avoid slowing down the productivity gains

A lawyer asks an AI for case citations supporting a specific legal argument. What is the risk, and what should they do?

There is no risk — AI was trained on legal databases and citations are reliably accurate The risk is low if the AI is asked to cite recent cases from 2024 or later High hallucination risk — AI may generate convincing-looking but nonexistent case citations. Every citation must be verified in a legal database before use The risk is only if the lawyer is in a jurisdiction not covered by US law training data

What is 'calibration failure' in the context of AI hallucination?

The model's expressed confidence doesn't match its actual accuracy — AI states invented information in the same confident tone as verified facts, giving users no signal that something is wrong When an AI's internal clock is set to the wrong time zone When an AI model is calibrated to a different language than the user's The process of adjusting an AI model to produce shorter responses

How did the relationship between users and AI change with generative AI tools like ChatGPT?

Users became more passive — generative AI automates everything without requiring input Users lost control — AI now makes all decisions without user input Users shifted from passive recipients of AI decisions (algorithms deciding what you see) to active operators directing AI to produce content and analysis on their behalf The relationship didn't change — generative AI works the same way as recommendation algorithms