Lesson 002Field guide · AI Basics
What AI Can and Can't Do
Set realistic expectations: AI's real strengths, its blind spots, and how to spot both.
You'll learn
- The pattern behind AI's real strengths
- The five failure modes to expect
- Why confidence tells you nothing about correctness
- The three-lane rule: use freely, verify, or decide yourself
- How to test an AI tool's reliability on your own expertise
01Section
Two things are true at once
By now you've probably seen both stories: AI writes a solid draft in ten seconds, and AI states something completely wrong without blinking. People tend to resolve this by picking a side — "it's brilliant" or "it's useless." Both camps get burned.
The truth is less dramatic. AI is genuinely strong at some kinds of work and unreliable at others — at the same time, in the same tool, sometimes in the same answer. Using it well doesn't mean trusting it or distrusting it. It means knowing which kind of work you just handed it.
The rule this lesson is built on
Confident is not the same as correct. An AI tool sounds exactly the same either way.
Checkpoint
02Section
What it's genuinely strong at
Intro to AI listed what AI is good at. Here's the pattern behind that list: AI shines when the job is transforming words you give it — and when you can judge the result yourself at a glance.
Summarizing
Long report in, short version out — and you can spot-check it against the original.
First drafts
Emails, posts, plans: a rough draft in seconds that you edit into shape.
Explaining
A concept broken down at whatever level you ask for — a patient tutor, on topics you can sanity-check as you go.
Reformatting
Notes into tables, bullets into paragraphs, formal into friendly.
Brainstorming
Twenty options in a minute. You throw most away — the two you keep were worth it.
Adjusting tone
Same message, different audience: shorter, warmer, more formal.
The common thread
In every one of these, you supply the material and you judge the output. The AI moves words around — nothing has to be taken on faith.
Checkpoint
03Section
Where it genuinely struggles
Flip that pattern and you get the weak spots: jobs that depend on facts the AI has to supply itself, or on precision you can't verify at a glance.
- It makes things up. Facts, statistics, quotes, links, book titles — generated with the same fluent confidence as everything else.
- It doesn't know your situation. Your customers, your files, yesterday's meeting — unless you say it, it isn't there.
- It slips on math and logic. Multi-step calculations and careful reasoning can quietly go wrong in the middle.
- It agrees too easily. Push back and it often folds — even when it was right the first time.
- Its knowledge has a cutoff date. Recent events, prices, and product versions may be outdated or guessed.
Plain-language definition
Hallucination — the industry's word for AI stating something false as if it were fact. Not a glitch, not a lie — just the pattern machine producing something likely-sounding instead of something true. Every AI tool does it, including the best ones.
Checkpoint
04Section
The confidence trap
Here's why this catches smart people. We're used to reading confidence as a signal: when a person answers fluently, with specifics, in complete sentences, it usually means they know the subject. AI breaks that signal — its tone is exactly as polished when it's wrong.
A fabricated answer, delivered perfectly (invented for this lesson)
"Yes — the 2019 Henderson study at Stanford followed 1,200 small businesses and found a 34% revenue increase. It's covered in chapter 3 of 'The Automation Advantage.'" Every detail here is the kind AI invents: a named study, a real-sounding university, precise numbers, a plausible book. Fluency proves nothing.
A quick self-check
Before repeating an AI-provided fact, ask yourself: would I bet $100 that this is true? If not, verify it first. The checking takes minutes — the wrong number in front of a client costs more.
Checkpoint
05Section
A simple decision rule
You don't need to re-evaluate trust from scratch every time. Before you hand AI a job, sort it into one of three lanes:
Use freely: Drafts, ideas, summaries, rewrites, explanations of things you can check. You're the editor — the worst case is a weak draft.
Verify before relying: Facts, numbers, names, dates, quotes, links — anything you'll repeat to someone else. Confirm it in an original source first.
Keep humans in charge: Legal, medical, financial, and people decisions. AI can inform the thinking; a person makes the call.
Key takeaway
Hand AI the drafting. Keep the verifying and the deciding.
Pause and think: Think of the task you most want AI's help with this week. Which lane is it in?
Checkpoint
Prompt exercise
Grade the AI on something you know
The fastest way to calibrate your trust is to test the AI where you're the expert. Copy this prompt into ChatGPT, Claude, Gemini, Copilot, or whichever AI tool you have access to — the website doesn't run AI itself. Then grade the answer like a teacher.
Ask me to name a topic I know very well — my job, a hobby, my hometown, anything. After I answer, explain that topic to a beginner: include specific details, the misconceptions people commonly have about it, and three practical tips. Be as specific as you can. When you're done, I'll tell you what you got right and what you got wrong.
Reflection: What was the mix — mostly right, a few fudged details, something invented outright? Remember that ratio. It's roughly what happens on topics where you can't spot the errors.
Quick check
3 quick questions — no pressure
There's no pass or fail here. Answer them all, and we'll show you the answers either way.