Meet the Machine

The idea

An AI assistant is an extremely well-read helper that predicts what a good answer looks like.

It has read a staggering amount of writing — books, manuals, letters, arguments, recipes, transcripts, encyclopedias. Out of all that reading it learned the shape of a good answer. When you ask it something, it is not walking to a filing cabinet and pulling out your file. It is composing, a few words at a time, the kind of answer that questions like yours tend to get.

That one sentence explains nearly everything else about its behavior, including the parts that seem strange.

What it is genuinely great at

• Writing and rewriting. Letters, notes, invitations, complaints, thank-yous, the difficult email you have been putting off.
• Shortening. Hand it four pages and ask for the point in five sentences.
• Explaining. It will explain the same thing five different ways without a flicker of impatience.
• Organizing. Give it a mess of notes and ask for a list, a table, or a schedule.
• Getting you off a blank page. First drafts are its best trick by a wide margin.
• Patience. It has never once sighed at a question.

Where it goes wrong

It is weakest on specifics: exact names, dates, prices, phone numbers, opening hours, page references, and anything very recent or very local. Ask for the hours of the hardware store in your town and you may get hours that sound entirely reasonable and are entirely invented.

Here is the part that matters most. It does not know when it is doing this. There is no change of tone, no pause, no hedge. A wrong answer arrives wearing exactly the same confident voice as a right one. Confidence is part of what a good answer looks like, so confidence is what it produces — whether or not the content behind it is sound.

Confidently making things up happens often enough that the tech world gave it a name: a “hallucination.” Now you know the lingo.

Why the same question gets two different answers

A small amount of randomness is built in deliberately. Without it, one question would produce one identical paragraph forever, and everything it wrote would read like a recording. The variation is why it sounds alive.

Two consequences for you. First, a different answer is not a malfunction and not a sign that one of them is a lie. Second, if an answer looks thin or odd, asking again is a perfectly reasonable first move — and asking the same question in a fresh conversation is a cheap way to see whether an answer holds up.

The analogy

Picture a brilliant new employee on their first morning. They have read every book in the building. They have never been to your town, never met your family, and have no idea what happened here last week. They are quick, eager, and desperate to be useful.

Ask them anything and they will answer — clearly, warmly, confidently, because that is what a good first day sounds like. Their confidence tells you about their manner. It tells you nothing whatsoever about whether they are right.

You already know how to work with this person. You have hired them before. You give them real work, you read what comes back, and you check the parts that would hurt if they were wrong. That is the entire job, and you have been doing it for thirty years.

Next: the single biggest thing you control — how you ask.