Lesson 0.4 · Tier 0 — AI Basics
Last lesson of the tier, and the shortest — because safety advice that takes an hour doesn’t get followed. Here’s everything you need on what never to paste into AI, built on one mental model: treat the chat box like a talkative, brilliant consultant in a café. Wonderful for thinking out loud. But you don’t slide your ID card across a café table, and you don’t read client files aloud where the neighboring table can hear.
Where does your text actually go? To the company’s servers, where — depending on your settings — it may be stored, reviewed for safety, or used to improve future models. Not a public billboard; not a diary with a lock either. That’s all the theory you need. Now the five categories.
The five never-paste categories
1 · Keys to accounts. Passwords, PIN codes, recovery phrases, banking credentials, API keys. There is no task that requires them — any prompt works with the password replaced by “mypassword.” If a website or “AI tool” asks for credentials to another service inside a chat, you’ve found a scam, not a feature.
2 · Identity numbers. Passport, national ID, full credit card numbers, bank accounts. Need help with a form? Paste the form with the numbers as XXXX — the intern fills in structure and wording; you fill in digits at home.
3 · Other people’s private information. The sneaky one, because the task feels innocent: “summarize this client contract,” “draft a reply to this patient,” “analyze this salary spreadsheet.” Your own secrets are yours to gamble; your client’s, patient’s, or employer’s are not — and depending on your work, pasting them may violate an NDA, professional duty, or privacy law before the AI even answers. This category is where careers get dented.
4 · Documents under confidentiality. Anything you signed an NDA over, unreleased company plans, legal disputes in progress. If leaking it to a stranger in a café would be a firing offense, the chat box counts as the stranger.
5 · Things you couldn’t bear to see again. The quiet category nobody lists: written in rage, deeply intimate, or about someone who didn’t consent to being material. Not because the machine judges — because copies outside your control are copies outside your control.
The redaction habit (keeps 100% of the usefulness)
Here’s the beautiful part: the intern almost never needs real names to do the work. Structure, tone, and logic survive redaction perfectly:
Paste: “Draft a payment reminder to [CLIENT] of [COMPANY] about invoice [NUMBER] for [AMOUNT], 30 days overdue, tone firm but preserving the relationship.”
Swap the placeholders back after copying the result. Ten extra seconds, zero lost quality, and every category above becomes workable instead of forbidden.
From the build log: the receipt that doesn’t exist
Every lesson in this tier has shown you a screenshot from this site’s build log. This one can’t — and that’s the receipt. Months of conversations built everything you’ve read: strategy, designs, budgets, even the AI’s own caught mistakes, all screenshotted and published without hesitation. What you will never find in that log, no matter how hard you scroll: a password, a bank number, an ID, a client’s name. Not because it was scrubbed afterward — because it was never typed. The hosting credentials that run this site have passed through zero chat windows. When money came up, it came up as decisions (“$5,100 — worth it?”), never as account numbers. The proof of good data hygiene is an absence — a build log so clean you can publish it, which is exactly what this academy does. That’s the standard for your build log too: keep it publishable from day one, and you’ll never face the choice between sharing your story and protecting your life.
One settings note before you graduate: every major tool has privacy controls — typically a toggle for whether your chats may train future models, and a way to delete history. Open your chosen tool’s settings tonight, find the data or privacy section, and set it the way you want. Two minutes, and you’ll know exactly what deal you’ve made.
Tier 0: complete
Take inventory of what you now carry: a working model of the machine (well-read intern, no shame) proven by receipts, an account you use daily, ten conversations that built real things, immunity to four myths — including one demonstrated live on this site’s author — the closing move, and the redaction habit. That’s more operational literacy than the vast majority of people arguing about AI online. All that’s left is the quest.
First: open your AI tool’s privacy settings and make your choice about training and history. Second: finish your Tier 0 quest — the one real problem, solved start to finish. You picked it in Lesson 0.0; if it’s still unsolved, this is the day.
Problem solved? Then it’s official: you’re no longer a beginner. Welcome to the lore.