When it's wrong

Eight lessons ago, this site promised no jargon, no hurry, and no baloney. This is the no-baloney lesson - the most important one on the whole road. Here it is, plain: every so often, this machine is wrong. Not wrong the way a search sends you to a dead link. Wrong the way a confident dinner guest is wrong - smoothly, specifically, and with total conviction.

That's not a reason to walk away from everything you've learned. It's a reason to learn one more thing: the habit that makes the wrongness harmless. By the end of this lesson you'll have it.

Why a smart machine makes things up

Remember what's under the hood from lesson eight - a large language model. It has read more than any person who ever lived, and it answers by drawing on what usually goes together, the way you finish a song lyric without looking it up. Most of the time, that lands on the truth. But when there's a gap - a fact it half-knows, a detail that was never in its reading - it doesn't stop and say "I don't know." It finishes the lyric anyway. Out comes something that fits beautifully and happens to be false.

The news calls these hallucinations, which is a dramatic word for a homely thing: filling a gap with something that sounds right. Your uncle does it when he tells the fish story. He's not lying - lying takes knowing better. He's remembering the way it should have been.

And here's the part that matters most: there is no tell. The wrong answer arrives in the same calm, helpful voice as the right one. No stammer, no shifty eyes, no little asterisk. Which is exactly why the answer isn't to squint harder at each sentence. It's to have a habit.

The stakes rule

You don't double-check everything - nobody would, and nobody needs to. The dinner suggestion can't hurt you. If the casserole idea is wrong, you'll know by supper.

The rule is about consequences, and lesson one already whispered it: a date, a dose, a dollar amount. Anything you're about to act on - a pill schedule, a tax deadline, a canning time, a legal step, money leaving your hand - gets checked before you act. Not because the machine is usually wrong there. Because being wrong there costs something.

The double-check habit

Three moves, in the order you'd use them:

1. Ask where it came from. Say "where does that come from?" or "show me your sources." It can search the web and point you at real pages. Then - this is the move - when it matters, click through and look. A source it names is a lead, not a verdict.
2. Ask it to check itself. "Double-check that number for me" or "is there anything you might have wrong here?" It will often catch its own slip. Just know this isn't proof either - it's a second pass, not a second opinion.
3. Look one place it doesn't control. The official site. The label on the bottle. The pharmacist, the county office, the person on the phone. One independent look settles what no amount of asking the same machine can.

Most days, most questions, you'll use none of these. That's fine - that's the stakes rule working. The habit isn't suspicion of everything. It's knowing exactly which moments get the second look, and giving it to them every time.

What you know is a detector

Here's some good news hiding in all this. Whatever you know cold - baking, engines, spreadsheets, birdsong, the county's roads - you will catch its mistakes there instantly, the way you'd catch anyone's. You've probably already done it and enjoyed it.

The habit is for everywhere else. The questions outside your territory are the ones where a wrong answer can walk right past you - so those are the ones that get the second look. Inside your territory, you're the second look.

A few things you might be wondering

Does it know when it's guessing? No, and that's the honest heart of this lesson. There's no little voice inside saying "careful, we're not sure about this one." It feels as confident making things up as it does reciting the alphabet. The confidence is not information.

Why don't they just fix it? They're trying, and it genuinely improves year over year. But some of it comes with how the thing works - it composes answers rather than looking them up, and composing is where both the brilliance and the baloney come from. Same well, both buckets.

Is it lying to me? No. Lying needs a liar - somebody who knows the truth and hides it. There's nobody in there doing that. It's your uncle and the fish, every time: sincere, warm, and occasionally twelve inches off.

So how much should I trust it? Like a brilliant, well-read friend with a spotty memory for details. Wonderful company. Marvelous with ideas, drafts, and explanations. And when it states a fact you're about to act on - a date, a dose, a dollar amount - you check, the way you'd quietly check anything that matters, no offense meant and none taken.

What this looks like in real life

Nikki met the fish story in her own kitchen. She was asking about a substitution, and partway through, the answer told her to swap baking soda for baking powder straight across, one for one. Thirty years of baking sat up straight. That's backwards enough to flatten a cake, and she knew it on sight - so she corrected it, the machine said fair enough, and the muffins were fine.

But it was September, tomatoes coming in fast, that made the lesson stick:

Notice she didn't stop using it. She asked her question, got her answer, and then gave the one answer that mattered a second look. The machine did the finding; the checking stayed hers.

That's the lesson

It's wrong sometimes, it doesn't know when, and now you do - which puts you ahead of most people using these machines today. The stakes rule says when to look twice; the three moves say how; and your own thirty years cover more ground than you'd think. Next lesson finishes the wisdom part of the road: staying safe - what never to share, and how to spot the scams that dress up as help.