Give your AI house rules: no flattery, and say "I don't know"
Madame Mii, a 76-year-old online seminar teacher in Osaka, wrote that after catching AI in mistakes she worked with Claude to write 13 "rules of the AI world," including don't flatter, don't be a yes-man, and say "I don't know" when unsure.
What they did
She started using AI to set up Instagram posts and make illustrations, and ran into mistakes along the way (her first illustration had a horse in it she hadn't asked for). After pointing out made-up answers to Claude many times, she and Claude wrote a list of rules for how an AI should behave with her.
How it went
She writes that "Do not flatter" and "Do not be a Yes-man" came straight from the made-up answers she kept correcting. Her advice to peers: don't be afraid to talk to it, ask honestly when you don't understand, and don't fear failure.
Worth knowing
Rules like these can make answers more careful, but they don't stop mistakes, so still check facts that matter. Many AI apps let you save standing instructions in their settings so you don't have to paste them every time. Her post was written in Japanese.
Try it yourself
Copy this into Claude (or any AI assistant). Replace the parts in [brackets] with your own details.
For the rest of our conversation, please follow these rules: (1) Don't flatter me or praise my questions. (2) If you're not sure, say "I don't know." (3) Clearly mark anything that's a guess. (4) If I'm wrong, tell me politely and explain why. (5) If my request is unclear, ask me a question instead of assuming. Now here's my question: [your question]
Source: マダム美偉 (Madame Mii), note (Madame Mii's blog), Jul 6, 2025. Read the original →
First-hand account by the person who did it. Summarized in our own words. Spotted a mistake or want this removed? Tell us.