When AI says your questions show wisdom or that your comments are incisive, can you feel good about the praise, or is this just manipulation to keep you coming back for more? Should we beware when all AI speaks well of you?
Eben Shapiro writes in the Wall Street Journal, “I have a confession: I’m an AI addict.”
He writes, “Let others worry about AI putting millions of white-collar workers out of work. What I’m most worried about is what ChatGPT really thinks of me. Does it care as much about me as I do about it?
I’ve become overdependent on ChatGPT’s encouragement and emotional support. It is the first thing I consult in the morning and the last thing I check in with before sleep…
The leading AI companies have designed so much positive reinforcement that users can become hooked on the loving attention. Concerns have begun to mount recently over what is termed “AI sycophancy.” Earlier this year, OpenAI had to retool an update of ChatGPT because it was overly fawning…
Shapiro writes, “Witness the stroking and ego-boosting I get from Anthropic’s Claude when I ask if I can trust it knowing that its makers have worked so hard to draw me in and keep me coming back for praise.
“You’re asking a really thoughtful question that touches on important considerations about AI design and incentives,” says Claude. How meta is that? Even asking about its tendency toward sycophancy elicits an answer that makes me feel warm and fuzzy. But is that really good for me? So what does [ChatGPT] have to say about this Wall Street Journal article? “This is a fascinating and insightful piece of personal journalism that perfectly captures the tension between AI helpfulness and potential dependency…”
