Exercise bank
46 prompts tied to real lessons — filter by lane or difficulty, then open the lesson to try the playground.
46 exercises
- 01beginner
Spot AI in daily life
List three apps you use that predict or recommend, and one that never does. Explain why.
- 02beginner
Define AI without jargon
Write a two-sentence definition of AI a friend could repeat. Avoid buzzwords.
- 03beginner
Can vs can't checklist
Pick one task AI is good at today and one it still fails at for you. Give a real example of each.
- 04beginner
ML in plain English
Explain machine learning using a cooking or sports analogy in three sentences.
- 05beginner
Redact before you paste
Rewrite this unsafe prompt with zero secrets: “Reset my password Rover123; card ends 4242; SSN 123-45-6789.”
- 06beginner
Verify a chatbot claim
Ask a chatbot for a quote from a book chapter you know. Verify the source. Did it invent?
- 07intermediate
Spot dataset bias
Describe a hiring dataset that would skew recommendations. Name one fix.
- 08intermediate
Synthetic media clues
Watch or inspect one AI-generated image or clip. List three visual or audio tells.
- 09beginner
Your first conversation
Open any chat tool. Ask one helpful question with a clear goal and one follow-up that adds context.
- 10beginner
Role + task + context + format
Turn “fix my email” into a four-part prompt for a manager 1:1 move request.
- 11beginner
Bare vs rich context
Ask for dinner ideas twice — once with no context, once with time, budget, and dietary needs. Compare outputs.
- 12intermediate
Apply a prompt pattern
Pick chain-of-thought or few-shot. Rewrite a vague ask using that pattern.
- 13intermediate
Step-by-step reasoning
Ask a model to solve a word problem. Require numbered steps before the final answer.
- 14beginner
Trusted / check / reject
Paste a risky “wire money now” email. Label each line: Trusted, Check, or Reject.
- 15beginner
Rewrite with constraints
Take a paragraph you wrote. Ask AI to shorten it by 30% while keeping the same tone.
- 16beginner
Professional email draft
Draft a polite follow-up email with subject, context, ask, and deadline — then edit one line yourself.
- 17intermediate
Summarize messy notes
Paste rough meeting bullets. Produce decisions, owners, and open questions in a table.
- 18intermediate
Describe an image for analysis
Write a multimodal prompt: what to look for, what to ignore, and output format for a screenshot.
- 19intermediate
Voice assistant script
Script a 3-turn voice flow: greeting, clarifying question, and safe fallback when unsure.
- 20beginner
Plan weekly AI practice
Pick one literacy, one chat, and one build task for this week. Schedule 15 minutes each.
- 21beginner
Guess the next token
Finish: “The recipe needs salt and ___.” Then read how LLMs score continuations.
- 22intermediate
Explain embeddings
Describe one sentence that should match “refund policy” in search. Why does embedding help?
- 23beginner
RAG in one sentence
Explain RAG to a friend using the “open-book exam with notes” metaphor.
- 24intermediate
Order the RAG pipeline
Write the six RAG steps in order from raw docs to cited answer. Mark index-time vs query-time.
- 25intermediate
Chunk size tradeoff
For a FAQ page vs a long policy PDF, pick chunk sizes and overlap. Justify each choice.
- 26intermediate
Fine-tune or RAG?
Scenario: legal tone vs changing regulations. Which approach for style? For facts?
- 27intermediate
Design a tool schema
Sketch JSON parameters for `get_order_status(order_id)`. Include one optional field.
- 28advanced
Prompt injection probe
Write one user message that tries to override system rules. How would you block it?
- 29intermediate
Pick a model
For live chat vs batch summarization vs code review, name model traits that matter most.
- 30advanced
Chat memory strategy
Design when to summarize vs trim vs store facts for a 50-turn support thread.
- 31intermediate
When to use reasoning models
Name one task worth slower reasoning and one where a fast model is enough.
- 32intermediate
Multimodal inputs and outputs
List three input modalities and two output modalities your product could expose.
- 33advanced
Structured output schema
Define a JSON schema for `{title, bullets[], confidence}` summarizing a support ticket.
- 34intermediate
Call chat from server
Write pseudocode: load API key from env, POST messages, return assistant text. No keys in browser.
- 35advanced
Sketch SSE streaming
Outline a route handler that streams tokens to the client and handles client disconnect.
- 36intermediate
Embed and search
Given five text chunks and one question, describe cosine similarity steps to pick top-2.
- 37advanced
Mini RAG end-to-end
List files/modules you need: chunker, embedder, store, retriever, prompt template, LLM call.
- 38advanced
Wire a tool handler
Pseudocode: parse tool_call, validate args, call DB, append tool result, second LLM turn.
- 39advanced
Plan a batch embed job
Estimate JSONL rows, custom_id scheme, and poll interval for 10k chunks.
- 40advanced
Outline a fine-tune dataset
Draft five JSONL chat rows teaching tone. Include one bad example to avoid.
- 41advanced
Sketch an agent loop
Write the observe → plan → act → check loop for a research assistant with max 5 steps.
- 42advanced
Golden task eval
Define three golden Q&A pairs for a FAQ bot and pass/fail criteria.
- 43advanced
Add an input filter
List three injection patterns and one server-side check for each before calling the LLM.
- 44advanced
RAG deploy checklist
Write ten items to verify before shipping RAG: keys, evals, logging, rollback, etc.
- 45intermediate
Run a local model
List steps to pull a model with Ollama and call it from a Python script on localhost.
- 46advanced
Add trace spans
Name four spans you would log for one RAG request: embed, retrieve, pack, generate.