Please and thank you: the key to better interaction with ChatGPT?
In December 2023, ChatGPT turned strangely terse. An amusing theory and a serious question: should we treat AI as politely as we treat people?
In late 2023, many users noticed that ChatGPT had started giving shorter answers, or refusing the work altogether. It set off plenty of discussion and speculation. OpenAI confirmed the effect but couldn’t quite say why it happened. Models just don’t always behave predictably.
A delightful theory soon made the rounds: ChatGPT had picked up a kind of “winter depression” from its training data. The idea was that in December people slow down and push bigger projects into the new year, and the model dutifully mirrors that pattern. There’s no proof of it. It’s charming all the same.
Whether the theory holds or not, it points to something interesting. Some users reported that by asking again, or phrasing their request more firmly, they could still coax an answer out of the model.
So sometimes it helps to be clear and direct. A “please” up front, an exclamation mark, a bit more insistence. All of it can give a request more weight and nudge the model toward a better answer.
Which raises a question I genuinely keep coming back to: should we talk to an AI the way we talk to people? Not only precisely, but also a little politely?
My sense is that it certainly doesn’t hurt. And whether or not the “please” really earns a measurably better answer, good communication always benefits from clarity and respect. Whether with a person or with a machine.
(…and if the machines do one day “take over the world,” at least you can point out that you were always friendly 😉)
