The AI Cheat Sheet: What Nobody Teaches You About Getting This Right
- Rich Washburn
- 54 minutes ago
- 8 min read
I've been at this for a while now. Long enough to have watched two very different groups of people encounter AI tools — one group that lights up and gets genuinely better at what they do, and another group that pokes at it for a week, concludes it's overhyped, and moves on.
The difference has almost nothing to do with which tool they're using. It has almost nothing to do with their technical background. It has everything to do with a handful of mental models that nobody bothers to explain.
This is the article I wish had existed when I started. Think of it as a cheat sheet — but the kind that actually sticks because it's built around how this thing works, not just a list of tricks.
Let's go.
First: What AI Actually Is (And Isn't)
Before anything else, you need the right mental model. Because if your mental model is wrong, you'll use the tool wrong, get mediocre results, and miss something genuinely useful.
Here's the most honest plain-English explanation I know: A large language model is a prediction engine trained on an enormous amount of human text. It has processed essentially the entire publicly available internet, plus books, papers, code, manuals, and more. When you type something to it, it's predicting: given everything that came before this, what should come next? That sounds simple. The emergent behavior from doing this at scale is extraordinary. But understanding the core mechanic helps you know both where it shines and where it breaks.
It's good at: language, logic, synthesis, summarization, brainstorming, explanation, rewriting, research framing, and pattern recognition in text.
It breaks down at: real-time information, precise math in some cases, things requiring external verification, and anything that happened after its training cutoff.
It is not a search engine. Google retrieves pages. ChatGPT synthesizes ideas. These are fundamentally different. One retrieves. The other thinks with you. Using it like a search engine — short, context-free queries — is the single most common mistake beginners make.
It is not an oracle. It can be confidently, fluently, convincingly wrong. The technical term is hallucination. It's gotten much less common, but it still happens. Never rely on AI output for medical, legal, or high-stakes decisions without independently verifying it.
It is not going to replace you. More precise: it won't replace people who know how to use it. Your judgment, your relationships, your experience — AI amplifies all of that. It doesn't replace it.
The One Shift That Changes Everything
Here it is. This is the whole game: Stop treating AI like a vending machine. Start treating it like a very smart, very fast colleague.
A vending machine: you put in a code, you get a snack. No context needed. No relationship required.
A colleague: you give them background, explain what you're trying to accomplish, have a back-and-forth, push back when something doesn't land, iterate until it's right.
Think about the difference between these two requests to an actual human colleague:
> "Write something about our new product."
vs.
> "Hey — I'm drafting the announcement email for the Q3 launch. Our audience is existing customers who've been waiting for this. I want it to feel excited but not hype-y. We're a B2B brand and our people hate fluff. Can you give me something under 150 words that leads with the problem it solves?"
The second one gets you something usable on the first pass. The same principle applies to AI. Context is the whole game.
The 4 Mental Skills That Separate Winners
These are the foundational habits. No tool will compensate for not having these.
1. Change your default reaction.
When you're stuck — on a question, a decision, a task, a concept — what's your first instinct? For most people, it's Google, or call somebody.
The shift: ask AI first. Not because AI always has the best answer, but because you'll get further, faster, and you'll show up to every follow-up conversation already informed.
This one habit, applied consistently, compounds faster than any new tool you could adopt.
2. Treat it like a new hire — not a search engine.
You wouldn't hand a new employee a task on day one with zero context and expect perfection. You'd brief them. Give them background. Correct them when they're off track. Build the relationship over time.
The people who declare AI useless are the ones who typed one vague prompt, got a mediocre result, and walked away. The people who win treat every session like an ongoing working relationship — because that's exactly what it is.
3. Build it continuously. Give it feedback. Make it better every day.
Your AI setup — your custom instructions, your memory files, your saved prompts — should get sharper every single day you use it. After a productive session, ask your AI to reflect on what worked and update its approach based on your conversation.
Most people never do this. The ones who do build a compounding advantage that's nearly impossible to close.
4. Stop trying to replace roles. Use AI to make yourself smarter in the room.
The question isn't: can AI replace my accountant?
The better question: can AI make me smart enough that my next meeting with my accountant is strategic instead of remedial?
Yes. Absolutely. That's where the real leverage is — not in replacing the expert, but in elevating yourself so you can operate at a higher level with every expert you work with. Lawyers, accountants, engineers, advisors. AI makes you a more informed principal in every one of those relationships.
The Six Building Blocks of a Great Prompt
Getting strong results isn't about memorizing magic phrases. It's about communicating intent clearly. There are six elements that make a prompt work. You won't always need all six — but knowing them changes every interaction.
1. Task — The non-negotiable. Start with a verb: write, analyze, summarize, rewrite, compare, explain. The verb tells the model what mode to operate in. - ❌ "Give me some marketing ideas." - ✅ "Generate five LinkedIn post concepts for a cybersecurity firm targeting mid-market CTOs."
2. Context — Just enough background to make your task specific. Who's the audience? What's the goal? What constraints matter?
3. Examples — If you have a sample of what you want, paste it in. Better yet, paste in your own writing and say: "Use this as my style reference. Write in my voice." The model stops sounding generic immediately.
4. Persona — Who should it respond as? A CFO reviewing financial risks thinks differently than a plain-English communicator explaining something to a curious teenager. Persona shapes vocabulary, assumptions, and angle.
5. Format — Tell it exactly how you want the output. Bullets, numbered list, short paragraphs, table, one sentence. If you don't specify, you get whatever the model decides.
6. Constraints — Length limits, tone guardrails, things to avoid. "Under 100 words." "No jargon." "Don't use the word 'leverage.'"
The shortcut: Task + Context + Format gets you 80% of the way there on most prompts. Add the others when the output needs to be precise.
10 Power Codes Worth Knowing
These aren't tricks. They're framing commands — ways of shifting how the model approaches your question. Drop any of these into a prompt and watch the output change.
Truth Mode — Stops ChatGPT from agreeing with you and forces honest, unfiltered feedback. Use this when you think you might be in an echo chamber.
80/20 — Asks for the highest-leverage path. "Give me the 20% of actions that will drive 80% of the result." Works for learning, strategy, skill-building — almost anything.
Unlearn — Surfaces outdated assumptions that might be holding you back. Especially useful in fast-moving fields where conventional wisdom ages poorly.
Socrates — My personal favorite. Instead of giving you answers, ChatGPT teaches you by asking questions one at a time. Forces active thinking. Builds real understanding instead of passive consumption.
Lindy Mode — Filters the response through what has worked for a long time, not what's currently trending. Named after the Lindy Effect — what survives tends to keep surviving. Great for strategy and life decisions.
Red Team — Argues against your idea. Shows you the weakest points in your thinking before you take it to the room. Use before any pitch, proposal, or major decision.
Pareto — Points you to the few inputs generating the biggest outputs. Cuts noise fast.
ELI10 (Explain Like I'm 10) — Forces plain-language simplicity. If the model can't explain it clearly, the response isn't actually clear.
Steelman — The opposite of Red Team. Builds the strongest possible case for your idea. Pressure-test from both directions.
Devil's Advocate — Pushes back on whatever you just said. Good for stress-testing any decision before you act on it.
How to Stop Writing Like a Robot
You can always tell when someone copied AI output directly into their content. The signals are everywhere once you see them:
- Em dashes used constantly — like this — and this one too — and this one - Colons introducing: every clause - Transition words: Moreover. Furthermore. It's worth noting that. In today's rapidly evolving landscape. - Anything that sounds like a business school case study wrote it
The fix isn't to avoid AI for writing. The fix is a cleanup pass. Save this prompt to your ChatGPT memory:
> "Act as a skilled editor. Revise the following AI-generated text to sound authentically human. Remove em dashes, excessive semicolons, and corporate transition phrases like 'moreover,' 'furthermore,' and 'it's worth noting.' Replace passive voice with direct, active sentences. Eliminate filler phrases that pad length without adding meaning. Keep the ideas intact — make it sound like a real person wrote it."
Use it every time. Thirty seconds. Removes 90% of the tells.
The Exercise You Should Do Right Now
Open ChatGPT. Free account is fine. Type this:
> "I'm going to give you context about me, and then I want you to write a short, conversational LinkedIn post about something I care about professionally. Here's the context: [your name], [your job title or what you do], [one thing you believe strongly in your field]. Write in first person, confident tone, no buzzwords, under 100 words."
Fill in the brackets. Hit enter. Read the result.
Now pick one thing that's slightly off — a word that doesn't sound like you, a claim that's not right, a tone that's too polished. Tell the AI what's wrong. Ask it to fix that one thing.
That's it. You just used AI as a collaborative tool. You gave it context. You pushed back. You iterated. That's not a beginner move — that's how it's supposed to work. And now you know the difference.
The Bottom Line
The AI gold rush is real. But most people are panning for gold in the wrong river.
The winners aren't the ones who tried every new tool as it dropped. They're the ones who built a real working relationship with AI — gave it context, trained it to their workflow, used it to elevate their thinking, and stayed consistent long after the novelty wore off.
Four mental skills. Six prompt components. Ten power codes. One editing prompt. That's the whole cheat sheet.
Save it. Use it. Come back to it in six months when the tool landscape has changed again — because none of this will have.
Rich Washburn is a technologist, strategist, and AI infrastructure builder. He works at the intersection of artificial intelligence, capital formation, and physical infrastructure through Eliakim Capital and Data Power Supply.