::AI literacy quiz · 5 levels · 30 questions

Where do you actually sit?

Honest calibration · no trick questions · explanations teach. Your answers save locally · no signup · no email · no tracker. Reset and retake anytime.

0 / 30 answered

Q1 · weight 1

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

Q2 · weight 1

What does a model's 'context window' mean?

Q3 · weight 2

You ask ChatGPT a question and it gives you a confident answer with three citations. What's the correct next step before relying on it?

Q4 · weight 2

Which of these is the WORST thing to paste into a public cloud AI chat (ChatGPT, Claude.ai, Gemini) without enterprise privacy settings?

Q5 · weight 2

What does RAG (Retrieval-Augmented Generation) actually do?

Q6 · weight 3

Your task is summarizing a 200-page PDF of confidential legal contracts. Which approach has the best privacy/quality tradeoff for a solo operator?

Q7 · weight 3

What's the practical difference between fine-tuning a model and using RAG?

Q8 · weight 3

What is an 'embedding' in AI systems?

Q9 · weight 4

What does MCP (Model Context Protocol) standardize?

Q10 · weight 3

An 'AI agent' is best understood as:

Q11 · weight 2

You're paying $20/month for ChatGPT Plus AND $20/month for Claude Pro AND $20/month for Gemini Advanced. What's the most common honest reason to do this?

Q12 · weight 2

BYO-key (Bring Your Own Key) means:

Q13 · weight 3

Which task is the BEST fit for a local model (e.g. Llama 3 on your laptop) instead of a frontier cloud model?

Q14 · weight 2

A 'hallucination' in LLM output is:

Q15 · weight 3

You ask a model to cite the case 'Smith v. Jones, 412 F.3d 891 (9th Cir. 2005)' and it confirms the citation exists and quotes the holding. What's the right move?

Q16 · weight 3

What does 'temperature' control in LLM sampling?

Q17 · weight 3

Which statement about API pricing is MOST accurate for typical frontier models in 2025-2026?

Q18 · weight 4

Prompt caching (offered by Anthropic, OpenAI, Google) does what?

Q19 · weight 3

What's a 'jailbreak' in the context of LLMs?

Q20 · weight 4

Prompt injection attacks target:

Q21 · weight 4

A 'mixture of experts' (MoE) model architecture means:

Q22 · weight 2

What does it mean that a model has a 'knowledge cutoff'?

Q23 · weight 3

Which is the BEST description of 'chain-of-thought' prompting?

Q24 · weight 4

An agentic system has 'read access to your email and write access to send emails.' What's the highest-risk failure mode?

Q25 · weight 4

You're choosing between Claude Opus, GPT-5, and Gemini 2.5 Pro for analyzing a 500-page contract. The deciding factor is most likely:

Q26 · weight 3

What's the honest reason most 'AI replaces my job' claims overshoot?

Q27 · weight 4

'Distillation' in AI means:

Q28 · weight 4

Which of these is a real, well-documented limit of current LLMs (2025-2026)?

Q29 · weight 3

Why does it matter that AI training data has a known geographic/linguistic bias?

Q30 · weight 3

You're building a customer support bot using RAG over your help docs. A user asks a question your docs don't cover. What's the safest behavior?

← back to /learn