Why ChatGPT is like false eyelashes

The one about a time I embarrassed myself

I’ve been seeing a lot of comments along the lines of “I just saw what AI can do, and trust me, it’s never going to replace writers. What a joke. We have nothing to worry about.”

And it reminds me of the time I said to my long-time hair dresser, “I would never wear false eyelashes. They look so fake. It’s ridiculous!”

And she stopped working on my hair and said, “I wear false eyelashes …”

I. Was. Mortified. But then she introduced me to a whole world of false eyelashes I had never known. I had no idea there are so many different kinds, and although they can be tricky to apply at first, once you get the hang of it, they can look great — just like well-applied mascara. (I’ve never gotten the hang of it.)

The lesson is that you only notice the bad implementations. I write about AI fails too, and it’s important to call out companies when they use it irresponsibly or stupidly, but when someone knows what they’re doing and makes choices that align with your tastes, you’re unlikely to notice they’ve used an “enhancing technology,” whether it’s eyelashes or an AI writing tool.

Tip: Keep a Prompt Journal

When you find a prompt that works, throw it into a document so you can use it again later. People call this a “prompt journal.”

On caveat I learned the hard way is that sometimes tools like ChatGPT work better when you have them do things in a step-wise way. For example, when I use it to clean up transcripts, it usually works better if I first have it do one thing (like put words used as words inside quotation marks) and then the next thing (fix capitalization problems), and so on. I’ve been more likely to encounter problems when I give all the instructions in one prompt.

Tip: Limit Your Custom GPTs

I still owe you the run-down on the new custom GPTs that let you train a model on your own data, but I want to quickly pass along a tip Leo Laporte from This Week in Tech gave me:

I was having a problem that when the custom GPT couldn’t find the answer to a question in my data, it would go out to the big general ChatGPT to get an answer, which wasn’t what I wanted. But Leo says you can actually tell the GPT to give you answers from only the material you’ve provided. Problem solved!

Erin Servais of AI for Editors had an interesting thought about custom GPTs too: She said maybe someday every book will be sold with a custom GPT that was trained on the book so you can ask it questions. I can imagine asking everything from “Geez, how many times did the author describe the hero as having ‘chisled abs’?” to “What was that anecdote about boomers taking phone calls at the table and millennials doing something different?” to “Give me a bullet list of every practical thing the book said you can do to overcome stage fright.”

What is AI sidequest?

Using AI isn’t my main job, and it probably isn’t yours either. I’m Mignon Fogarty, and Grammar Girl is my main gig, but I haven’t seen a technology this transformative since the development of the internet, and I want to learn about it. I bet you do too.

So here we are! Sidequesting together.

If you like the newsletter, please share it with a friend.

* No news today. I shouldn’t have even spent time writing this abbreviated newsletter! I’m supposed to be working on my book launch, and my LinkedIn Learning courses, and my podcast, and, and, and … but I have so many thoughts and tips I can’t stop, and every time, I think, “I have just this one little thing to share. It won’t take long.” Ack!

Written by a human. Copyright 2023, Mignon Fogarty, Inc.