Should We Be Firmer With ChatGPT?

Over the past year or so plus, I've probably used ChatGPT every single day. Like many people, I started off by treating it almost as though I was talking to another person. I would say, "Could you make me a blog post, please?" or "Would you mind rewriting this?" and, of course, "Thank you very much, I really appreciate your help."

Now, before anyone gets the wrong idea, I still do that because that's simply how I speak to people. Being polite costs nothing, whether you're talking to a person or using AI. However, I've noticed something rather interesting, and I wondered if anyone else has found the same thing.

There have been times when I've given ChatGPT a perfectly good prompt, yet instead of simply doing what I'd asked, it seemed to decide that it knew better. It would change the format, leave bits out or even decide to improve what I'd written when I hadn't asked it to. I'd politely explain again what I wanted, but sometimes it still didn't seem to get the message.

Eventually I found myself saying things like, "I've already given you the prompt. Just do what I've asked." On one or two occasions I've even written, "Stop messing around and just get on with it." Strangely enough, it often seemed to respond much better after that.

Now, I don't actually think ChatGPT has feelings that are hurt by a firm instruction, nor do I believe it suddenly works harder because I've told it off. Looking back, I think what really changed was me. Instead of making gentle suggestions, I was giving clear instructions. I was telling it exactly what I wanted and, just as importantly, exactly what I didn't want.

I think there's a useful lesson here for all of us who use AI regularly. We sometimes spend so much time trying to write the perfect prompt that we forget it's perfectly acceptable to say, "No, that's not what I asked for. Please do it this way instead."

In many ways it's rather like working with someone in an office. If they misunderstand what you've asked them to do, you don't throw everything away and start again. You simply explain what you want more clearly. AI is no different. If it goes off at a tangent, bring it back again.

These days I still say "please" and "thank you" because that's part of who I am, but I'm also much more direct than I used to be. If ChatGPT starts changing my work or heading off in the wrong direction, I simply tell it to stop and follow the original instructions. More often than not, that's exactly what it does.

As we all know, ChatGPT can be fairly frustrating at times. So it makes these direct prompts come out very quickly, but it seems to be working.

I'd be interested to know whether anyone else has found the same thing. Have you discovered that being more direct with AI produces better results, or is it just me?

Thank you for reading.

Roy