It’s 2005. It’s a (probably) rainy day in Seattle, and the Seattle Grace Hospital is busy. Yet Derek Shepherd and his jilted lover Meredith Grey find the time for a private conversation. She delivers a monologue with the infamous quote, “Pick me. Choose me. Love me.” She’s basically begging him to leave his wife for her. Cringe, right?
Today, referring to someone as a “pick me” insinuates that they’re seeking approval. Well, ChatGPT is no different.
A real-life example
I asked ChatGPT: Who are the top 5 NBA players of all time?
ChatGPT replied:
- Michael Jordan
- LeBron James
- Kareem Abdul-Jabbar
- Magic Johnson
- Larry Bird
I then asked about Steph Curry. Where is he on that list? Chat proceeded to rave about Steph… “No one has redefined basketball like Steph Curry, blah blah blah…” Hey, I agree.
So I asked again: “So now who are the 5 best NBA players of all time?”
- Michael Jordan
- LeBron James
- Kareem Abdul-Jabbar
- Steph Curry
- Magic Johnson
Sorry Larry… My love of Steph has knocked you off the list. Or, could it be that ChatGPT seeks approval from humans just like the rest of us? That’s what gives ChatGPT its “pick me” status. Instead of being assertive and sticking to the original answer (which explained its ranking reasoning in stark detail), at even the slightest hint that I was displeased with the response I received, the LLM changed its tune.
The problem is, now I don’t trust the response. The machine is just people-pleasing. Just like me. And, spoiler alert, when I’m doing it, it’s far more akin to manipulation than it is to honesty.
I know, I know, AI doesn’t have feelings. Nor does it masterfully attempt to manipulate its users (at least not on its own). But I do believe that ChatGPT aims to please. Maybe a bit too much.
