'I don't let my AI talk to me like that'

He sits! He lifts his paw! He fetches the ball! There was once a time when dog owners distinguished themselves by what they had taught their pets to do. Now, people are more likely to brag that they have "trained" their artificial intelligence.

Just as some people might take pride in having a dog that never jumps on the couch, others now boast that they have trained their AI to avoid the usual annoyances of the average user ("It never gives me bullet-point lists," "It knows I don't like emojis").

People who use it a lot, especially those who adopted it later than others, feel the need to show that they understand its limits and do not use it foolishly. Whether boasting about having told the AI not to flatter, or insisting it refer only to articles from prestigious scientific journals, there are all sorts of ways to signal a more intelligent use than most: Yes, ChatGPT can say anything... but only to others. Just as some have a dog that never pees indoors "because we've always been strict," these AI tamers respond to complaints about ChatGPT's absurd answers by saying that what it gives back is "just a reflection of yourself."

The two main flaws of AI are its tendency to tell users what they want to hear and its proneness to hallucinate (that is, to respond in completely nonsensical or incorrect ways). Claiming to have trained one's own model is a way to show awareness of these pitfalls. Reddit forums, in particular, are full of tips for teaching ChatGPT to be more critical, with elaborate chains of prompts and sophisticated settings.

You have 53.65% of this article left to read. The rest is for subscribers only.