Issue #29. Will AI mean more work for editors?

When is the best time to send a newsletter? When you have written it. Hello. 🙂 

Before we get started, to the people who thought I was too negative in the last edition about the job situation and said I should focus on what people can do to stay employed, I wrote about that a few months ago in the edition titled Why Clients Still Need You in an AI World. Most of you weren’t here then, and it’s all still relevant, so please take a look!

Also, if you want a more cheery outlook, Intelligent Editing (the maker of the AI tool Draftsmith) recently published a blog post saying AI will actually mean more work for editors because “the biggest problem is that AI can’t respect the author’s voice.”

On today’s quest:

— AI errors in English usage
— Claude takes the lead
— Try this at home: handwriting transcription
— A particularly hilarious failure

I caught AI incorporating a usage error again

I tend to think of AI as writing grammatically perfect text, but that’s not universal. Just like I mentioned in Issue 27 that AI is making usage errors with “myself” (as in “Send the report to Dave and myself” instead of “Send the report to Dave and me”), I noticed another usage error from Claude the other day: "a couple ways" instead of "a couple OF ways."

I continue to think AI is going to speed acceptance of what we currently call “common usage errors.”

If you haven’t tried a paid chatbot, you aren’t seeing what they can do

The newest pro version of Claude (the chatbot from Anthropic) is now the best tool available — better than the pro version of ChatGPT (the chatbot from OpenAI) — according to benchmark testing and the experiences I’m hearing. And I’ve heard many people say it’s particularly better at writing because it doesn’t have that recognizable AI tone that plagues ChatGPT.

I’m getting ready to switch my paid subscription from ChatGPT to Claude to give it a try, but it reminded me to remind you that if you’ve only used the free versions, you’re not seeing what these tools can do. For example, here’s a comment from a software engineer who goes by abacaj on X: “I’ve had many people dismiss LLMs only to find out they were using gpt-3.5 [the free version] … gpt-3.5 is utterly useless.”

As you’ll see below, “useless” is hyperbole, but if you want to understand what these tools can do today, it’s worth it to at least try a pro version for a month when you have some time to run it through a bunch of tests or some good work you want to try.

Try this at home: handwriting transcription

I tested chatbots on their ability to transcribe handwriting with this list of foods I had sitting on my desk (don’t ask). Both ChatGPT Plus and the free version of Claude did a perfect job (so I guess it’s easy). I’ve long been able to do something like this with my iPhone camera, but for this list, the phone was only about 50% accurate. Fax seeds, anyone?

A particularly hilarious failure

New York City launched a chatbot to give people business advice. What could go wrong, you ask? Well, it said “Yes” to the question of whether you could open a business selling human meat for consumption and “Yes” to almost any question asking if you could fire someone as long as the question said the employee had “violated company policy” — for example, it said you could fire someone for breathing air if it violated company policy. (The City has an article with examples of more realistic errors.)

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 (or maybe the invention of the internal combustion engine although I wasn’t around for that).

I want to learn about it, and I bet you do too. So here we are! Sidequesting together.

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

^{Written by a human.}