As Far as You Know, an AI Didn’t Write This

Friends,

As is the case in the majority of the non United States parts of the world, last week was May the 1st, or Labor Day. Here in Taiwan, it’s an official government holiday. However, it doesn’t cover public servants, including teachers and doctors, but does include other people working at schools. So that day, my TA gets the day off, our department is closed, but classes continue. With plenty of absences, though: I assume because their other classes were cancelled, or maybe because their friends/family had the day off for work and they didn’t want to miss out.

Fun fact, not only did the US make sure Labor Day wasn’t May 1 so it couldn’t be linked to international solidarity or an already politically significant day, they even went as far as to make May 1 Loyalty and Law Day. I am not kidding.

Students in the US are finishing up their semesters even as we speak, but National Taiwan University is beginning Week 12 of 16. A lot of professors hate the end of the semester. It’s a mountain of grading and a bunch of people who were ambivalent about their grade suddenly deciding to care. I find those negatives are balanced out by everyone being in a good mood, contemplating the light at the end of the tunnel. That said, I still haven’t figured out what my medical class’s final is going to be (more specifically than a presentation) so… we’ll see how my perception adjusts over the next month.

I’m signed up for another year at NTU, and in fact, my department is proposing a whole new course that, if approved, I’ll be teaching. It’s on using AI-based tools to assist with writing. There’s no guarantee it will be approved. While the university as a whole has put out positive statements about encouraging the use of these tools, I’m sure there are many professors who would be against such a course. In addition, even if it is approved, is it something students would want to take? Would they trust a writing professor to teach them better than online tutorials about a new technology, to the extent that they would give up an elective slot?

There are big pros to teaching such a class. It’s new, which is by itself exciting, but also means it could lead to some easy articles about the possible pedagogy. It’s also clear that AI writing tools are going to affect my career in the future no matter what I teach, so facing it head on will be informative and useful in that regard. Students should be driven and curious to learn all they can.

It also has cons, of course. There’s not going to be a wealth of previous lesson plans to help me decide what to teach. I have a bunch of ideas, but I’m not sure I have anywhere near sixteen weeks of ideas. And what happens if ChatGPT goes paid users-only mid semester? That’s pretty unlikely to have a big effect, given the existence of Bard and a few others, but having a course completely beholden on a private company’s free tool seems like a bad idea.

I think what I’m most interested in about the course is that I’m not sure how I feel about AI as a writing tool yet! My views are still flexible; I find myself believing slightly differently as I continue to learn. For example, I was very much anti-AI art and AI-writing for journalism/other copy (I specify these because they’re the first mainstream uses that it does decent with, so far) at first. I still definitely am, but I’ve read several compelling arguments that our moral panic over AI-generated art while accepting automation in nearly every other industry as being progress has a tinge of classism. Obviously, art for art’s sake is incredibly important. But there’s also a whole lot of art for a paycheck’s sake, like it or not, and haven’t I previously argued that in those situations we should be treating writing as less precious and more pragmatic than we currently do? Also, how do I account for procedurally-generated content, like Sigur Rós’s Route One ? Or even the maps inside a video game?

Honestly, I have no idea. And it seems like it’s not going to matter, because it seems very clear that we’re soon going to be unable to differentiate between human- and AI-generated art, in which case “catching” people using it will be even more impossible than it already practically is. But the fact that I don’t know fully how I feel about it makes it more interesting. I want to read more, I want to have more conversations, I want to try it out myself.

Right now, it can’t do the assignments I ask my students do. It’s really good at summarizing data. It’s very bad at explaining where it got that data. If you ask it to read an article, it will usually do a good job. If you ask it to interpret some conclusions from that article or combine two articles’ information, it will make things up out of left field. Complex writing, from a formal abstract for an already written article to a story in the style of someone, are rarely close to their mark. But, in some cases, that’s just a matter of giving it better instructions.

Not always. I had a long, weird conversation with both Bard and ChatGPT where I’d ask it to write a poem in the style of Emily Dickinson about a random topic, in this case, carrying laundry when it’s hot outside. They could get the vocabulary close, but the structure would be the kind of poems we read in elementary school, where every line is the same length and it’s all AABB rhyming. Then I would point out the problem with the style and the AI would agree with me, agree that it should be different, but then when asked to make a new poem using this knowledge, it would produce another of the same vein.

—
“write a poem in the style of emily dickinson about how awful it is to have to carry laundry to the laundromat when it’s grossly hot outside”

The sun beats down upon my back
As I trudge down the dusty track,
A heavy burden in my arms,
A load of laundry, full of charms.

Oh, how I long for a cool breeze
To ease the sweat that starts to tease
My brow, my neck, my tired frame,
As I carry on, in heat and shame.

The laundromat looms in the distance,
A haven from this scorching existence,
But still I struggle, with each step,
Through waves of heat, I must be kept.

Why must this chore fall on me,
To carry clothes, so endlessly,
In heat that saps my very soul,
And makes me wish I’d never strolled?

But onward still, I must press on,
Until this task is fully done,
And then I’ll rest, and hope for rain,
To wash away this heat and pain.
—

Maybe I’ll be retired before it can actually do these things well. Probably not.

A couple of months ago I included a link to Vanderbilt’s EDI deans using ChatGPT to write an email to students about a shooting. I actually think this is a pretty good use case for the tool. In what world is a dean going to have an actually meaningful message to tell students in this situation, unless they’re willing to go totally outside the box? Maybe it’s more an indictment that an email like that is meaningless. The deans, in this case, were reinstated soon after . I’m curious what my students will have to say.

And I’m curious what you have to say, too! Send me thoughts! Send me articles you found interesting. Have you tried using it? My doctor students say they’ve saved a lot of time on some emails. Maybe I should’ve asked it to write this TinyLetter! I’ll try it in the future, see if anybody notices.

Further reading:

• Finally found a new vegetarian broth that I’m in love with! Just One Cookbook ’s vegetarian ramen needs a couple of tweaks (I’d recommend subbing in a little coconut milk), but it is so good for how simple it is.
• For the last month I’ve had variations on playlists built on top of “Last Days” by Yasushi Yoshida playing. It’s melancholy piano music, perfect for long stretches of work, and I love it so much, but I honestly have no idea how I started listening to it. If you recommended it to me, let me know?
• Behind the scenes at Taco Bell! Food science! Food architecture!
• Didn’t find this profile of dril to be particularly interesting or worth reading. I would, in fact, call it an insufferable and boring article. Despite that, I’m sharing it because I can’t believe dril is younger than me. He feels like an immutable corner of the internet that’s been there since time immemorial.
• Admired this example of a meaningful, accomplishable qualitative research project: bathroom quality across campus .
• Was unaware that tiny Japanese trucks are being imported into America for farmers , but it makes sense, since the default gigantic truck is not practical for most uses. Now I want an in-depth analysis as to why most work trucks in Taipei (mainly used to deliver ingredients and disturbing numbers of gas canisters to restaurants) are all the same color of blue.
• Come for the scathing portrait of the Goop Cruise, stay for the deserved swipes at DFW and his readers (incl. myself!).
• Have long been obsessed with the idea of a VH1 Where Are They Now -style investigation into random people from our past. I hope it’s a crucial part of the afterlife debriefing room. Monica Potts did something similar for this article (and her memoir): trying to understand why she left her small town in Arkansas, why others stayed, and what happened to them all.
• Favorite Twitter follow of the month: Mondo Mascots , which chronicles the multitudes of mascots from Japan, as well as occasionally from other places.
• Finally, was very thankful for Michelle Kuo’s writing on continuing to learn Mandarin . She’s a few steps ahead of me, but it’s interesting where our experiences with language learning overlap and diverge. Plus it’s just such good writing.

For the past month I’ve joined a second, daytime Chinese class twice a week. It’s frustrating, because it’s reminding me just how much I’m forgetting of what I’m learning. Very, very much two steps forward, one step backwards. But it’s also a clear reminder of how just taking more steps in general is the only way forward. Sounds cheesy? Oh well. I hope you find yourself comforted by experiencing the truth of a cheesy statement, as well.

-g