First Exercises
Three exercises built on the parts from the last chapter. They take half an hour and need nothing but an open window.
Exercise 1. Just ask
Start with something you already know. The Eiffel Tower, Lisbon, Mozart: any landmark, any city, anyone out of a school textbook.
The first question is short:
Then keep asking. Who built it and with whose money, what stood there before, why it was going to be torn down 20 years later.
Every answer pulls the next question out of you. Working with AI runs like a conversation, and a single prompt rarely finishes the job.
Exercise 2. Plan a trip
Pick a country or a state you have been meaning to visit. A short opening line will not do here: everything the model would otherwise drag out of you with a dozen small questions, say at once.
Dates, the arrival and departure cities, the number of people, and the budget are the frame of the trip. Without them you get somebody else's plan.
Name the budget in your own currency; there is nothing to convert. Portugal pays in euros, and the model will do the math.
The exchange rate it uses is approximate and can be off by a few percent. For a rough estimate that is enough; for a calculation it is not.
Name your preferences just as directly, and say right away what matters more. "We like museums" says nothing: Lisbon alone has dozens.
Ask for a day-by-day plan. Read it. Say the plan is too packed and that you will not survive that much on day two, and ask for a rewrite.
You have just done what you will be doing every day: refusing to accept the first answer.
Exercise 3. Choose by criteria
Take a purchase you have coming up: a car, a television, a washing machine.
Say everything that matters up front:
Do not name models. Name conditions. How much you are ready to spend, where it will stand, who will use it, what matters most to you.
Ask for options that fit those conditions and for the reasoning behind each one.
Now add yourself. The apartment is rented, the machine stands in the kitchen, a child sleeps on the other side of the wall. Or the room is narrow and you mostly watch sports.
Push back. Say that noise matters more to you than spin speed, and watch the list change.
Some of the specs from the first answer will drop out, and some small detail will come out on top. That is the point of the exercise: you work out which specs are yours and which belong to someone else.
The model knows about products what made it into training and what it could reach on the internet. Both can be out of date.
Take the specs from there, but check the price and whether the product is even sold at the store. While you are at it, ask outright what year the data is from.
What the three exercises show
The model is fast. When it lacks data, it will sooner produce a plausible answer than admit it does not know. It depends on how you worded the task. And it knows nothing about your project, your company, or your boss (what a wonderful person he is).
Three exercises took you half an hour, and that is enough to stop being afraid of the blinking cursor. The next chapter is about where the model goes wrong, and why.