Toolkit

The practical end of the site. A card to print and keep by the keyboard, and the hands-on tools from the lessons, gathered in one place. Everything here runs in your browser and nothing you type is sent anywhere.

The pocket card

What it is doing

Predicting plausible text, one piece at a time. It does not look things up unless the tool fetches something. One answer is a sample, not its view.

Four cheap checks

Ask for the quote it is relying on.
Ask twice, in a new chat.
Check the one key fact.
Better still, bring the facts yourself.

Before you trust a task to it

Is there one answer that must be exactly right?
Does it need information from outside the model?
Would I be able to tell if it were wrong?

Yes, yes, no: expect a cliff.

Ask like a form

Say what the output looks like. Give one example. Say what it is for. When it misses, fix the prompt and run it fresh.

Before you paste

No passwords or card numbers. Nothing identifying about other people. Nothing confidential from work. Strip the names; the advice will be as good.

Tone is not evidence

Confidence, structure, numbers and hedges are all style. Folding when you push back is not confirmation. If it were wrong, when would you find out?

Build a specification

Fill in as many of the boxes as you like. The prompt assembles itself as you type, and you can copy it straight into whichever tool you use. Nothing you type leaves this page.

The three-question test

Type a task you are thinking of handing to a chatbot, answer three questions about it, and see roughly where it sits on the map. Nothing you type leaves this page.

If this were wrong, when would I find out?

Pick the answer that fits what you are about to use a chatbot for. You get the check that is worth its cost, and no more.

Strip it before you paste it

Paste in something you were about to send to a chatbot. This finds the details that usually add nothing to the request: email addresses, phone numbers, long numbers, dates and names. It runs only in your browser, so nothing is sent anywhere. It will not catch everything, so read the result before you use it.

Your week of trying things

The ten things from above, as a checklist. Ticks are kept in this browser only, so you can come back to it. Print it if you would rather have it on paper.

Every lesson's move, in one list

The small action at the end of each lesson. Doing them is what makes the lessons stick; reading them is not.

What you are looking at. Open any chatbot you can use and type something with no goal at all. Ask it what it is. Ask it something silly. You are not trying to get anything done yet. You are just getting used to the empty box.
Where the words come from. Ask a chatbot for a fact you can check in ten seconds: a date, a population, the name of a street near you. Then check it. You are not testing how clever it is. You are finding out how much checking you need to do.
Never the same answer twice. Take a question you have already asked a chatbot. Open a brand new chat and ask it again, word for word. Compare the two. Whatever you find is worth knowing before you rely on either answer.
Why it invents things. Ask for three sources on a topic you know a little about, then try to find them. Do this once, on purpose, on a question that does not matter. It is much better to meet an invented citation when you are looking for one.
Confidence is a style. Take an answer a chatbot gave you recently and mark, honestly, which parts you actually checked. For most people, the first time they try this, the answer is none of it. That is the useful thing to notice.
How much it holds. In a long chat you already have, ask it to repeat something specific you said at the very start. Whether it can tells you where the edge of the window is for that tool.
Three kinds of memory. Start a new chat and ask it: what do you remember about me? Then find the memory screen in your tool's settings and read what it has saved. Delete anything you would not want carried into every future chat.
What it can reach. Ask the tool you use what today's date is, and whether it can open a web page. Two questions, fifteen seconds, and you will know more about your tool's limits than most of the people using it.
Writing like it knows. Notice one moment today when the chat pulls a social response out of you: you thank it, you apologise, you feel judged, you feel it is being lazy. You do not have to stop. Just catch it happening.
Good at that, bad at this. Pick a task you think it should ace, and one that feels almost the same but needs an exact answer. Run both. The gap between them is what to remember, not either result on its own.
Asking as specification. Take a prompt that gave you a disappointing result. Do not explain yourself to it. Add one example of what you wanted to the original prompt, and run it again from scratch.
When it declines. Next time you hit a refusal on something reasonable, rewrite the request to say who you are and what you need it for, then send it in a fresh chat. See what changes.
Cheap verification. Choose one of the four. Just one. Use it on your next real question, and keep using it until it is automatic before you add another.
What happens to what you type. Write your own one-line rule for what you will not paste in. Make it about your own life, not general. Then go and find those two settings.
Nothing here breaks. Ask it something you had assumed was too strange, too small, or too obvious to ask. That hesitation is the last thing between you and getting real use out of this.