Before Sunday: what to do, and ten minutes to read

What an AI assistant actually is

An AI assistant such as Claude is a program that has read an enormous amount of text and learned the patterns in it. You write to it in everyday language, and it writes back something that fits the pattern of a helpful answer.

That is useful, and it is also the reason to be careful. It does not look things up the way a search engine does, it does not know you, and it can sound completely sure while being wrong. Think of a quick, well-read helper who has never met you and sometimes guesses. You are still the one in charge of the work.

Good at, and careful with

• Good at: explaining something in simpler words, turning rough notes into a clean draft, suggesting options when you are stuck, summarising a long text you paste in, checking your writing for clarity.
• Good at: playing a role so you can practise, such as an interviewer, a customer, or a tough reviewer.
• Careful with: facts, figures, dates and quotes. Check anything that matters against a real source.
• Careful with: anything personal or confidential (see below), and anything where a wrong answer has real consequences, such as health, legal or money decisions.

Five words you will hear on the day

• Prompt: what you type to the assistant. A good prompt is mostly a clear request.
• Model: the AI system underneath. Claude is one; there are others.
• Context: everything the assistant can see in this conversation, including what you pasted in. It does not remember you between separate conversations unless a tool is set up to.
• Hallucination: when an assistant states something false in a confident tone. The fix is to check, not to trust the tone.
• Data privacy: what happens to what you type. Treat the chat as something you would be comfortable having a stranger read.

How to ask well: four parts

Most weak answers come from thin requests. A request with four parts works far better than a one-liner.

• Who you are and the situation: “I am a final-year student applying for a first job in marketing.”
• The task: “Rewrite the summary at the top of my resume.”
• What good looks like: “Three sentences, plain words, no buzzwords, confident but honest.”
• Your limits: “Do not invent any experience I have not told you about.”

Keep these out of the chat

• Passwords, one-time codes, card or bank details, and anything with an Aadhaar, PAN or passport number.
• Other people's private details: their numbers, addresses, health, or anything they told you in confidence.
• Confidential documents from a job or college that you are not allowed to share.
• When in doubt, change the names and numbers first, or leave it out.

Ask the assistant to explain a topic you know well, in three sentences. Then mark what it got right, what it got wrong, and what it left out. This is the quickest way to learn how far to trust it.

Paste a short paragraph you wrote and ask for three specific suggestions to improve it. Accept one, reject two, and say why. The point is to practise being the editor, not the audience.

Ask the same question in two separate conversations and compare the answers. You will see how much of the answer is steady and how much changes, which is exactly why you check.