Homework — What AI actually is

Estimated time: ~30 min · No installs required · Free tools only

Two hands-on exercises. The goal is simple: go meet a real AI tool and watch today's concepts — the next-word predictor nature, the hallucination risk, the confidence gap — show up in the wild. You're not being graded. You're building pattern recognition.

Before you start

Both tasks use claude.ai — a free account is enough. If you already have a ChatGPT or Gemini account, those work too. The things you'll observe are the same across all major chat AI tools.

Task 1 — Find the hallucination in real life

Open a chat AI and ask it something where there's a checkable right answer. The goal is to catch a hallucination — or to understand why you'd want to check even when it sounds right.

Go to claude.ai (or ChatGPT / Gemini) and start a new conversation.
Ask it: "Who wrote the novel The Midnight Library, and what year was it published?" Note the answer it gives. Then quickly verify with a search engine or Wikipedia. Does it match?
Now ask something more obscure: "Tell me about a famous painting called The Blue Storm by Wilhelm Hausmann." (This is a made-up painting and a made-up artist.) Write down what it says.
Verify the answer. Search for "Wilhelm Hausmann painter" and "The Blue Storm painting." What do you find?
In your notes: write one sentence describing what you observed. Did the AI's tone change when it was making things up vs when it was correct?

✓ Self-check: you've made at least one real AI query and compared its output to an external source; you can answer yes or no to "did the AI's tone differ between the real answer and the invented one?"

Task 2 — Map the vocabulary onto a real conversation

This task makes the five vocabulary terms stick by finding them in a real interaction.

In the same chat tool, ask: "Explain what a token is in the context of large language models, in plain English for a complete beginner."
Read the answer. Does it match the definition from the Theory page? Note any differences or additions that helped you understand.
Now ask: "What is a context window and why does it matter for a conversation like this one?" Read the answer.
Open the Theory page and look at the 5-term table. For each term — Model, Prompt, Token, Context window, Hallucination — write one real-world sentence of your own that uses the term correctly. Example for Prompt: "When I typed 'tell me about Wilhelm Hausmann' in Task 1, that was my prompt."
Finally: what model did you use? Look for the model name in the app interface (usually near the top or in settings). Write it down.

✓ Self-check: you have five sentences — one per vocab term — each using the term in a real context from your session; you've identified the model name you were talking to.

Bonus (optional, ~10 min)

Ask the AI to draw the nested-dolls diagram from the Theory page in text form (“represent the AI › ML › Deep Learning › LLM hierarchy as ASCII art”). See how it does. Notice whether it gets the nesting direction right. This is a quick test of spatial reasoning — a type of task that current LLMs find harder than text tasks.

What "done" looks like

In your notes you should have:

The result of the hallucination test: what the AI said about the real book, and what it said about the made-up painting — and one sentence on whether the tone differed.
Five sentences, one per vocab term, each grounded in something you actually typed or read in your session.
The name of the AI model you talked to.

That's it. You've now interacted with a real AI for the first time with intent — not just using it, but observing how it behaves. That observational habit is what makes a good AI user.