You have probably already got one of these open somewhere right now, haven't you? Maybe it is ChatGPT, or maybe it is whatever you personally landed on.

And you have used it. Maybe you use it every single day. But for one thing, right? A question here, an email there, something quick. It works, mostly, and truth be told you have never had a real reason to think there was any more to it than that.

But do you know what it can actually do beyond that one thing? Have you ever stopped and wondered whether you are using just a tiny fraction of something much bigger? Or did you just quietly assume that this is basically all there is?

Now let me tell you something about myself up front, because it matters more than anything else I could tell you. I am not a coder. I run one of these tools every single day, for real work, and I still remember exactly what it felt like to type something in and get back nothing worth keeping. That was not very long ago, and I have not forgotten it. So hear me on this: I am not writing to you from three years ahead of you, looking back down. I am writing to you from close enough that I can still see you from right here.

So picture a building with me. And not a small one either. Floor after floor of things this thing can do, more than you would believe if I stood here and listed them all out right now. But here is the thing. You did not walk into that building and get turned away at the door. You walked into the lobby, you got exactly what you came for, and you had no reason in the world to think there was an elevator.

And nothing about that was wrong. The lobby did its job just fine. It is just that the lobby is not the whole building, and nobody ever walked you past it. You cannot want what you have never been shown exists.

Now I am not going to teach you everything on every single floor. That is not the promise, and I am not going to sit here and pretend that it is. What I am going to do is walk you up, floor by floor, and show you exactly what is up there, in plain language, with nothing you cannot go and try yourself the same day you read it.

So if it is not me, then what is it?

So if it is not you, then what is it?

Let me just tell you straight, because the straight version is stranger than anything I could dress it up as. This is not a metaphor. A metaphor holds one thing up next to another so you can see it a little better, and I am not doing that here. I am just going to state what is true, and the truth is strange enough on its own.

It has read more than any person could get through in a thousand lifetimes. It has read more about your work, your questions, your industry than you could get through if you did nothing else for the rest of your life.

And it remembers none of it.

Not the way you mean when you say remember. Left to itself, it does not know your name, or what you asked it yesterday, or a single thing about you, until something puts that in front of it. And it cannot point back to where it learned anything, so it cannot tell you when it is guessing and when it truly knows.

It has read everything, and it remembers nothing.

That one sentence explains almost everything strange about the way this thing behaves, so stick with me here. Before this chapter is done, you and I are going to trace four completely different things it does right back to that same line.

Let me get one objection out of the way first, because you may already be forming it, especially if you have used these tools for a while. Your version has probably started remembering you, your name, your work, the way you like things, and here I am telling you it remembers nothing. So which is it? Both are true, and honestly the gap between them is the most useful thing in this whole book, so stay with me. What remembers you is a setting, switched on right next to the tool, that quietly keeps a few notes about you and hands them back before you ever type a word. That is real, it is often on by default now, and a later chapter is all about turning it to your advantage on purpose. But pull those notes away and the thing underneath still reads everything and does not know you from a stranger. It is not really remembering you. Something else is reminding it, every single time, from scratch. Keep that difference in your pocket, because everything in this chapter is true whether your version remembers you or not.

WHY IT CAN SOUND BRILLIANT

Ask it something you have been chewing on for months. Not a fact question, a real one. How do you word the conversation you have been avoiding with your business partner? Is the number the contractor quoted you actually reasonable?

What comes back is often better than you expected, and here is why. It is not digging through its own small pile of experience for something useful to say, the way a friend would. It has sat in on a version of your exact question thousands of times, across thousands of people who wrote about it, argued about it, and lived it, in books and forums you will never lay eyes on. So when it answers, it is drawing on all of that at once, instantly, and it never gets tired or bored halfway through.

Marcus, who has run his shop alone for six years, was deciding whether to take on a partner. He laid the real situation out, what the partner would bring, and what he was afraid of losing. He typed something close to this:

What came back started right where he expected, the money, the workload, the arithmetic of splitting a business in half. But a few lines in, it said something he had not thought to say himself.

It named the thing sitting underneath his hesitation, that he was not afraid of splitting the money, he was afraid of losing the one thing that made the shop feel like his. He told me it felt like being seen by someone who had sat across the table from a hundred people in his exact spot. And in a way, it had.

Think about it. Very few people you know have read that widely on the exact thing you are asking about. That is not hype, that is just what happens when something has read everything.

WHY IT CAN SOUND CONFIDENTLY WRONG

Now turn the coin over, because this side matters even more, and it is the part that gets people hurt.

Ask it something with a specific, checkable answer. What year did a rule change? What does a policy actually say? What is the number buried in a contract? It will often hand you an answer that sounds exactly as confident as the brilliant one you just read. Clear, specific, not a flicker of hesitation in it.

And sometimes it is simply wrong.

It is not lying to you. Lying means you know the truth and choose something else. What it is doing is reaching for the shape of a right sounding answer, the way you can finish the chorus of a song you have not heard in ten years without remembering who sang it or whether you have got the words exactly right. Most of the time the shape is close enough. Once in a while it is confidently, completely off, and no little warning light comes on to tell you which one just happened.

ASK IT TWICE, GET TWO ANSWERS

Here is a small thing that unsettles people the first time they catch it happening. Ask it the same question twice, a few minutes apart, and you can get two different answers. Not opposite, usually, but different enough to make you wonder which one is real. Nothing is broken. It is the same fact you already know, just wearing a new hat. It does not keep a fixed record that it reads back to you. It builds each answer fresh, from scratch, reaching again for the shape of a good response, and the shape it grabs is not identical every time. Two people asking get two slightly different replies for the very same reason one person asking twice does. Sofia, comparing notes with her sister, found they had each been handed a different figure for the same simple thing, and each assumed the other one had typed it wrong. Neither had. For most of what you do, this will not matter at all. But when a specific number or date actually counts, do not lean your whole weight on one phrasing of it. Pin the fact down, ask it plainly where that came from, and check the version that matters against something outside the tool.

Picture yourself asking it the return window on something you bought. It answers instantly, a clean specific sentence that sounds like it was read straight off a receipt. Sometimes it was. Sometimes it was stitched together from the shape of a thousand return policies, and not one of them was yours. From the outside those two look identical. The only way to tell them apart is to check, and almost nobody checks, because nothing about the way it was said gave them a reason to.

Beth, an attorney, found this out the hard way while prepping to speak at a town meeting. She asked it for a statistic so she would sound sharp on a topic she had not had time to research, she repeated it confidently in the room, and she got corrected on the spot. Not a disaster. Just the small, avoidable stumble that the accuracy check exists to spare you.

This is not some defect they forgot to fix. It is the direct price of the very thing that makes it brilliant. Something that has read everything and remembers nothing cannot tell you which of its million sources had it right, because it never kept the receipt. The brilliance and the confident wrongness are the same feature seen from two sides. You do not get the one without the other being possible.

One honest update for you, because the newer tools have genuinely gotten better at this. Ask a strong current model something shaky and it will sometimes flag its own uncertainty without you asking. That is real progress, and I want to give it its due. But it does not do it reliably, and the one moment it matters most, a specific number you are about to repeat out loud in front of someone, is the exact moment you cannot afford to learn afterward that this was one of the times it stayed quiet. The accuracy check is what covers you for the times it does.

WHY THE SAME QUESTION GETS A GENERIC ANSWER

Here is a third thing that same sentence explains, and I would bet you have lived it yourself without ever putting a name to it.

Two people type the exact same six words on the same day. What does liability mean here. One is a nurse asking about a form at work. The other one runs a small business and is asking about a client contract. They need completely different answers. The nurse needs to know what she is on the hook for if a patient gets hurt. The business owner needs to know what happens if a client sues.

But if neither one says who they are, both get roughly the same answer. Reasonable, generic, technically about liability, and useful to nobody in particular, because it was built to be a little bit useful to both at once. It has read everything there is about liability, for nurses and business owners and a hundred situations you have not even thought of. It remembers nothing about which one you are standing in, unless you go ahead and tell it. So it hedges. It hands you the version that is least wrong for the most people, which is just another way of saying the version that is most right for no one.

That is not a flaw you have to work around forever. It is a door that has been sitting unlocked the whole time, and nobody ever told you to push on it. The nurse and the business owner get two sharply different, genuinely useful answers out of the same tool in the same minute, the moment either one says a single sentence about who they are and what is at stake. And that is not some far off chapter. That is the very next one.

WHY THE CEILING IS HIGHER THAN YOU THINK

A fourth thing now, and this is the one people almost never think to ask about.

Everything it read was not just emails and articles. It read the manuals too, how software gets built, how files get organized, how a spreadsheet formula actually works, how a folder full of documents gets sorted, everything anyone ever bothered to write down about how they did it. It has read everything. Not everything except the technical parts, everything.

Have you ever asked it to actually do something with a file on your own computer, not just talk about one? Most people never have. Not because it cannot, but because the thought simply never crossed their mind, the same way it never occurred to Carol, the bookkeeper, that it could read a PDF, even after eight months of using it every single day for something smaller. Nobody ever said the sentence out loud to her, and you cannot want a thing you do not know is there.

The same tool that just explained your mortgage to you in plain words has, somewhere in everything it read, soaked up how a person with no technical background at all can describe what they want done to their own files and have it done. That is the same tool, just further up the building than most people ever walk. There is a whole chapter on it waiting for you later.

ONE MORE THING THAT CONFUSES PEOPLE

You may have noticed that a conversation gets better the longer it runs. You correct it once and the next few answers land closer to home. That is real, and it looks for all the world like the tool is learning you. It is not, though, not quite. Everything you say stays sitting in front of it for the rest of that one conversation, the way you still remember the start of a meeting by the time it wraps up. But close the window, open a new one, and the conversation itself is just gone. If your version has memory switched on, it will still greet you knowing a few saved facts about you, the ones we just talked about. But the twenty minutes you spent getting it to understand your full situation do not come back. That is not a glitch. Set the memory feature to one side and what you have is simply a thing with no memory of its own, doing exactly what such a thing does. There is a chapter later on how to fix that for good.

THE OTHER KIND OF FORGETTING, INSIDE ONE CHAT

A meeting has its limit too, though, and so does this. Julie, who plans events for a living, spent forty minutes in a single conversation building up her whole picture, the venue, the budget, the two dates she could not move. An hour later, still in that very same window, she asked it to draft the invite and it quietly used a date she had already ruled out near the top. It had not jumped over to a new chat. It had simply let the earliest part of a very long conversation slide out of view, the way the first ten minutes of a three hour meeting go fuzzy on you by the end. This is a different kind of forgetting than closing the window. It happens inside one chat, when the thread runs long, and the things you said earliest are the ones most likely to fade. The fix is small. When a conversation has wandered a long way from where it started, or you are turning to a genuinely new topic, open a fresh window and paste the handful of facts that still matter right back in. You are not starting over. You are just handing it the notes again.

TRY THIS

Do this one right now, it only takes a minute. Open up whatever tool you use. Ask it something you would ask a knowledgeable friend, something with real stakes for you. Read the answer, and notice how confident it sounds. Then, in that same window, ask it one thing you already know the factual answer to, a date, a number, something you could check in ten seconds. Read that one. Notice that it sounds exactly as confident as the first.

That is the whole chapter in one minute, felt instead of explained. One answer came from having read more than anyone alive on the subject. The other came from that same place, except this time the shape it grabbed was slightly off. And nothing in the way either one was said would have tipped you off to which was which.

THE LIGHT SWITCH

Here is the shift I want you to make before you turn the page.

Up to now you have been treating the flat answers, the wrong ones, and the generic ones as three separate annoyances. They are really one problem seen from three angles, and every one of those angles traces straight back to the same fact you already know. Once you truly believe it, you stop expecting it to know you on its own, the same way you would stop expecting a brilliant stranger at a party to know your kids names. And you start doing the one thing that fixes most of this whole book at once, telling it who you are before you ask it for anything that matters.

That is a light switch nobody ever showed you was right there on the wall. Once you know it is there, you flip it every single time without even thinking, and the difference in what comes back stops being subtle.

One honest limit before we move on. Knowing what this thing is will not, all by itself, get you better answers. That is a separate skill, and it happens to be the easiest one in the whole book.

So how do I actually ask?