Thuta Learning
AdvancedAIbeginner

Don't Ask for Secrets

Relax. We'll talk through this in plain words — no textbook voice.

What you'll walk away with

  • Understand Don't Ask for Secrets without it feeling intimidating
  • Get hands-on practice using it yourself
  • Learn to spot and sidestep invented facts and secret-sharing, with a smile

Take a breath — this chapter is written to be read like a conversation, not a textbook, so it's easy to follow.

Let's Think About It for a Second

The RAG course said a file's secrets can leak back out. A prompt is even more direct — whatever you hand over, it can remember, echo, or guess from. A student's grades, a family phone number, an API key, a password, someone else's exam — don't add those as context. Sending someone else's message and asking it to "fix it for me" isn't safe either. Don't hand the waiter your house key.

Connect It to Everyday Life

Fine to use — your own fake notes, classroom content, made-up names. Not fine — .env files, a class roster, medical records, someone else's message. Keep a constraint line in your prompt pack: "no secrets." Need an example? Use a fake name like Alex or Sam. You don't need a real name.

Let's Try It Together

text
Safe notes:   plants use sunlight to make food
Unsafe notes: API_KEY=sk-live-...  Alex scored 38/100  09-...
You should see
You can tell what shouldn't go in a prompt.

Try It in 5 Minutes

Write 6 lines of fake context. Split them into 3 "okay to use" and 3 "not okay," with why.

A Quick Heads-Up

Even a deleted chat sometimes doesn't fully disappear. Not sending it is the safest option.

Easy traps

  • Assuming "only my friend will see this"
  • Sending a screenshot without blurring out the numbers

Try It Yourself Now

Write 6 lines of fake context. Split them into 3 "okay to use" and 3 "not okay," with why.

You'll know it worked when: You can tell what shouldn't go in a prompt.