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

What is the '80/20 trap' in AI adoption?

AI tools cost 80% more than traditional software and deliver only 20% of the value AI can only be used for 20% of any professional's tasks; 80% must remain fully human Only 20% of AI tools are reliable; 80% are experimental and should not be used at work AI handles 80% of a task well but fails at the remaining 20% in ways requiring human correction — making autonomous completion risky, but AI-assisted draft work highly valuable

How quickly did ChatGPT reach 1 million users after its release in November 2022?

ChatGPT has never reached 1 million users — it is primarily used by researchers 1 year — mainstream adoption in consumer AI has historically been slow 6 months — it grew slowly at first before viral spread 5 days — the fastest product in history to that point to cross 1 million users

What does the history of AI 'winters' and breakthrough cycles tell us about the current AI era?

The history of AI winters is irrelevant — current AI is categorically different from all prior AI Because we've had AI winters before, today's AI is definitely not as advanced as claimed AI winters prove that AI never delivers; the current excitement will also disappear AI has had repeated cycles of hype and disappointment — the current era may be different in scale, but calibrated skepticism and attention to real results (not promises) remains the right posture

What are 'parameters' in a large language model?

The rules that prevent AI from producing harmful content The settings menu where users configure how the AI responds The physical memory chips inside the servers that run the AI model Billions of numerical values adjusted during training that encode the model's learned patterns — they represent what the model 'knows' about language, facts, and reasoning

Even the most capable AI models in 2026 cannot do which of the following?

Reliably verify whether their own output is factually correct — models can produce false information with the same confidence as true information Produce grammatically correct text in English Write code that executes correctly in common programming languages Summarize a long document into key bullet points

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

Delegating an entire client project to an AI agent with no human review of outputs Using AI to replace all written communication with a client without any human involvement 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 RLHF (Reinforcement Learning from Human Feedback) and why did it matter for ChatGPT?

RLHF is a hardware upgrade that made ChatGPT run faster than previous models A fine-tuning technique where human raters scored model responses, training the model to produce outputs humans preferred — turning a text completer into a helpful assistant RLHF prevents AI from using any data it scraped without permission RLHF is the process of connecting AI to the live internet for real-time information

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

Why is 'Artificial Intelligence' considered a misleading name by many researchers?

AI systems don't think or understand — they perform sophisticated pattern recognition and statistical prediction, which produces human-seeming outputs without human-like comprehension There is no controversy about the name — it accurately describes how the systems work AI is misleading because the system is not artificial — it uses real human data The name is misleading because AI is more intelligent than humans in every domain

What does a 'context window' determine in an AI conversation?

The number of different topics an AI can discuss in a single session The time window during which a conversation is saved before being deleted The size of the screen where AI output is displayed How much text the AI can 'see' and refer to at once — including your messages, shared documents, and the AI's own replies. Text outside the window is effectively forgotten

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

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

What usage pattern consistently predicts high-value AI adoption?

Using AI to replace human involvement in all written tasks entirely Starting with a specific, recurring real problem; using AI for first drafts with human review; building task-specific prompting habits; knowing which tasks to keep fully human Waiting for AI tools to mature further before integrating them into real workflows Trying every new AI tool as soon as it launches to stay current

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

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

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

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 is a type of machine learning, which is a type of AI — they are nested categories, not separate things Generative AI came before machine learning, which came before AI

What fundamentally changed about AI in late 2022 with ChatGPT's release, compared to AI that existed before?

ChatGPT was the first AI to run on computers; previous AI required specialized hardware AI was invented in 2022 — there was no meaningful AI before ChatGPT Generative AI crossed a quality and accessibility threshold — anyone could use it for genuinely useful work, without technical expertise, for the first time The 2022 breakthrough was AI becoming conscious and able to truly understand language

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 Most eventually reach a 'plateau of productivity' where real, consistent value is delivered — the trough represents reality calibration, not permanent failure 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

Why does an AI model perform better on common tasks than on rare or highly specialized knowledge?

AI models are intentionally limited on specialized topics to avoid replacing experts Common tasks appear more frequently in training data, so the model has seen more examples to learn from — rare or specialized topics have fewer training examples, increasing error and hallucination risk Common tasks are hard-coded into AI models; rare tasks require internet access Rare topics require more parameters, and models have a fixed number available

Which of the following tasks is AI MOST reliable at?

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

What is the core training objective of a large language model?

Create a database of question-answer pairs to look up when asked something Predict what word (token) comes next, given all the words before it — repeated billions of times across vast text to build pattern knowledge Build a logical reasoning engine that can prove mathematical theorems Memorize every fact in its training data so it can retrieve them accurately

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 Higher temperature always produces better quality — you should maximize it for all tasks 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 Temperature controls language formality: low for formal writing, high for casual writing