AI: What to Trust (and What Not To)
Let me tell you the single most important thing to know about AI chatbots, the thing I wish came printed on the box:
They are never unsure out loud.
Ask ChatGPT or Claude something it knows cold, and you get a confident, polished answer. Ask it something it doesn’t know, and you get… a confident, polished answer. Same tone. Same tidy paragraphs. That’s the whole problem, and once you understand it, you can use these tools safely for the rest of your life.
“Hallucinations,” explained like a human
When an AI makes something up, the industry calls it a hallucination. Fancy word, simple idea.
Remember, a chatbot works by predicting what words should come next, based on patterns from everything it read during training. It’s like a friend who’s read a thousand mystery novels: ask them how a mystery usually ends, and they’ll nail it. Ask them how a specific book they never read ends, and they might describe an ending that sounds exactly right — plausible detective, plausible twist — and is completely invented.
The AI isn’t lying. Lying requires knowing the truth and dodging it. The AI is doing what it always does — producing the most plausible-sounding text — and sometimes plausible and true part ways.
I taught myself to code decades ago, when there was nobody to ask and the only way forward was trial and error. That left me with a habit: never trust a tool until you’ve found its breaking point. So when chatbots arrived, I went hunting. I asked one for a biography of an obscure historical figure and got a lovely writeup — dates, a university, a famous mentor. About a third of it was fiction, woven seamlessly into the facts. I’ve since earned an AI certification through Google and Coursera, and nothing in the coursework changed that first lesson: the made-up parts don’t glow red. They sound exactly like the true parts.
The verify-before-you-rely rule
Here’s my rule, and it’s short enough to remember at the keyboard:
The more it matters, the more you check.
Brainstorming dinner ideas? Check nothing. Worst case, you eat a weird casserole.
Getting a plain-English explanation of how something works? Skim a reputable source if it matters.
A fact you’ll repeat, publish, or act on — a law, a dosage, a tax rule, a quote, a phone number, a price? Verify it independently before you rely on it. Look it up at the actual source: the government site, the manufacturer, the doctor.
One habit that helps: ask the AI, “How confident are you, and what should I double-check?” It’s often refreshingly candid about its own weak spots. Another: ask for sources, then actually click them. If a link is dead or the article doesn’t say what the AI claimed — and this happens — you’ve caught a hallucination in the act.
A client of mine once asked a chatbot for the hours and phone number of a local government office. Beautifully formatted answer. Wrong number, wrong hours. Thirty seconds on the official website would have saved a wasted trip. Now it does.
Where AI shines
Trust it enthusiastically for work where you remain the judge of the result:
Drafting and rewriting. Emails, letters, announcements. You’ll read it before sending, so mistakes can’t sneak past.
Explaining concepts. How mortgages, wills, or Wi-Fi work in general. Big-picture understanding is its home turf.
Brainstorming and organizing. Ideas, outlines, packing lists, meal plans. There’s no “wrong” to hallucinate.
Summarizing text you provide. When you paste in the document, it’s working from your material, not its imagination — though it can still oversimplify, so skim anything important.
Where it fibs
Keep one eyebrow raised for:
Specific facts: names, dates, statistics, prices, phone numbers, addresses.
Quotes and citations. It will invent scholarly-looking references with real-sounding authors. This one fools smart people constantly.
Current events. Its knowledge has a cutoff date, and unless it explicitly searched the web, “recent” answers may be stale or invented.
Math and legal/medical specifics. Fine for general understanding; not a substitute for a calculator, an accountant, or your doctor.
Confidence, calibrated
Golf taught me that confidence and accuracy are different sports. Every golfer I know — myself included — has stood over a putt certain it breaks left, watched it break right, and blamed the green. Chatbots putt with that same confidence.
You don’t need to fear these tools, and you shouldn’t worship them either. I’ve been in tech for 30-plus years and lived through enough sure-thing fads to know real usefulness when I see it — and this is real. Treat the AI like a brilliant, well-read friend who never admits uncertainty: wonderful company, great first draft, terrible sole source.
Enjoy the shine. Verify the facts. That’s the whole game.
And if you’d like hands-on help building these habits — on your own devices, at your own pace — send me your questions. I’d be happy to walk through it with you.