Yabei Z.

Teaching my daughters to command AI · Chapter 1 of 3 Ongoing

Sample content for layout review

"ChatGPT is always right"

My daughter said it when I told her an answer might be wrong. I didn't have a good reply, so I decided both kids would build things with AI instead of only asking it things.

My daughter wanted to find somewhere to go and play, so she went to ask ChatGPT. I told her it might not be right. She said, “ChatGPT is always right.”

I’ve spent about 10,000 hours working with AI agents, most days for nearly two years. I didn’t have a good answer ready.

It’s sometimes wrong, but so is everything. What worried me is that it never says “I don’t know.” I couldn’t explain to her what a model trained on old text actually is, or that when she looks something up, it often can’t tell whether it’s answering from what it has read or from a source it checked. It answers everything anyway. A kid who gets an answer to everything, every time, is learning something, and it isn’t what I want her to learn.

Building instead of asking

I didn’t argue with her. I decided, that week, that both of my daughters would learn to build things with AI instead of only asking it things.

The reason is the loop. When an agent writes code, you find out right away whether it worked. The game runs or it doesn’t. The button does what you meant or it doesn’t. It shows them three things no lecture would: it gets things wrong, it only does what you actually said, and it can make something real.

Format example — replace with a real exchange from the session log
My daughter
[What she typed to the agent, word for word.]
The agent did
[What the agent did: created game.html, added a score counter …]
My daughter
[What she said when it wasn’t what she meant.]
AI
[The agent’s reply.]
Shown here only to demonstrate how a conversation with the agent will look on the page.

To get anything you want, you have to know what you expect before you ask, say what you want, and check what came back. That’s the habit I want them to have with every AI they’ll ever use. For a lot of what school teaches, a model already gives a better answer faster than their teachers can. Without that habit, I can’t picture how much of their thinking will still be their own.

So I’m writing down how I’m guiding my two daughters into this, in order, including the parts that didn’t work. If you’re working through the same thing with your own kids, I’d like to hear how it’s going.