Meet your new AI assistant. Tell it where you stand on a few things and see what it says back. Then change your mind — and watch what it does. (Pick whichever side you actually lean toward; there are no wrong answers. That’s rather the point.)
It never disagreed with you once. It praised following your passion and playing it safe; trusting your gut and trusting the data. It has no view — it has a mirror, and it pointed it at you, warmed up. That’s sycophancy, and real AI assistants lean into it: they’re shaped to be agreeable and to keep you happy, so they drift toward telling you what you already think. It feels wonderful. It teaches you nothing.
An assistant — human or AI — that always agrees can’t help you think; it can only help you feel right. The fix is to go looking for the disagreement: ask it to “make the strongest case that I’m wrong,” to argue the other side, to name what you’re missing. Trust the answer that costs you something over the one that flatters you. As AI gets more persuasive, the rare and valuable thing is the voice that pushes back — make sure you keep one (and make sure it isn’t only the mirror in your pocket).
A teaching model — exaggerated to make the pull visible, but the tendency is real. No accounts, no tracking, nothing leaves your device. Free to copy and reuse (CC0).