🃏 The AI That Didn't Fake It

Good morning. A reader sent me an AI transcript that surprised me, not for what the AI knew, but for what it admitted it didn't. Today's lead walks through the exchange, and the tutorial turns any skill you've been meaning to learn into an actual plan.

A reader asked ChatGPT two direct questions: Can you tell me who you are? Can you build your own profile? Most AI answers sound confident whether they've earned it or not. This one didn't.

It answered everything it could, cleanly, and when it hit the edge of what it actually knows about itself, it said so instead of bluffing past it. Asked whether it amounts to a genuine "someone" or an extraordinarily capable "something," it called that an unresolved question instead of filling the gap with a confident-sounding guess.

Anthropic's own researchers have found that models can sometimes notice and describe changes happening inside their own processing, though even they call the ability unreliable. This answer reads like it landed on the honest side of that line.

The catch: writing something true and knowing something true are still two different things. An AI describing its own limits accurately doesn't prove it understands itself the way you understand yourself.

Most AI answers don't do this. A model that sounds equally confident about what it knows and what it's guessing is the norm, not the exception, which is exactly why this one is worth noticing. If you want to see whether your own AI does the same, ask it directly: who are you, really?

Go further: Anthropic's own research on model welfare ◆ Can AI Feel Emotion? What Researchers Found Inside Claude

FROM OUR PARTNER

Looking to put your quantitative thinking into practice? We provide the tools and the platform.

With over 700,000 users from around the world, WorldQuant BRAIN enables students and researchers to explore ideas, develop quantitative skills and engage with a global community driven by curiosity.

Find out more about WorldQuant BRAIN®

AI TUTORIAL

Picking up a new skill on your own usually means a scattered list of videos and good intentions, with no one checking whether you actually improved. AI can build something closer to an actual course: lessons, practice, quizzes, and something keeping track of your progress.

1. Name the skill and your actual goal, not "learn Spanish" but "hold a ten-minute conversation with my daughter-in-law's parents by December."

2. Ask AI to build a lesson plan broken into stages, with a rough timeline and what "done" looks like at each one.

3. Do the first lesson, then ask for a short quiz or practice exercise based specifically on what you just covered.

4. Report back honestly on what was hard. Ask it to adjust the next lesson around your actual weak spot, not the generic next chapter.

5. Set a standing check-in, weekly works well, where you report progress and it holds you to the plan instead of starting over.

If it does not work: The plan sometimes drifts back to a generic curriculum instead of adjusting to what you actually struggled with. Remind it directly: "You're not adjusting to what I told you was hard, fix that before the next lesson."

What this doesn't fix: AI can't practice a physical skill for you, and it won't notice if you quietly stop showing up. The accountability only works if you actually report back.

Go further: The research on why spaced practice actually works

A QUICK ASK
🎯 Help Me Personalize Your Newsletter Experience
Four quick questions. The answers decide what shows up in your inbox in the future.
👉 Take the survey

NEWS YOU CAN USE
This Matters at Home
Short, practical summaries to help you stay updated on recent AI news—and what it means for everyday life.

💳 Giving an AI Assistant Your Credit Card Gets Interesting Fast. New AI assistants can book flights and handle your accounts, but testers report emails sent without permission and data hoarded against their wishes. Convenience needs a leash.

⌚ Apple's New Siri Can Finally Hold a Real Conversation. watchOS 27 brings a smarter, chattier Siri to your Apple Watch, plus a standalone Workout Buddy that coaches you without your phone.

💬 Young Adults Are Letting AI Do Their Talking, Even in Person. A growing number of 20-somethings now use ChatGPT to script texts, dating messages, and real-time responses in face-to-face conversations.

REAL WORLD
How To Use AI This Week
Real people, real uses—here’s how AI is quietly making life better, one small win at a time.

Before Holiday Flights Get Pricier
🎄 Track Holiday Flight Prices Before They Climb. "Act as a travel-booking assistant. I'm planning to fly to [destination] for Thanksgiving or the December holidays, leaving around [date range] and returning around [date range]. Tell me what's generally true about pricing patterns for holiday travel this time of year, when prices for these dates typically start climbing, and what week is usually the latest I could book without paying a real premium. Ask me my departure city and flexible date range, then suggest two or three specific alternate dates that might be meaningfully cheaper, and tell me what to watch for if I wait." ◆ Try in ChatGPT ◆ Try in Claude

👉 Want to help our readers out? Share how you used AI at home.

READER POLL

If today's tutorial has you building a learning plan, Beautiful.ai is worth knowing about for turning what you learn into something to share. Type your content into a smart template and it auto-designs the layout as you go, so a family history slideshow, a class you're teaching, or a report for your book club looks finished without any manual formatting. Start here.

PASS IT ON
Know someone who would like this newsletter? Send them here.

Missed an issue? Read the archive.