Now that it’s here, I think I feel the same way about AI as most everyone else does. In the macro, I wanted to make our lives easier and not kill us all. In the micro, I want it to make my job easier but not actually take my job.
I'm a late adopter who's married to, and works with, an early adopter. So when ChatGPT first became available and they all said AI was SPECIFICALLY going to decimate people in my line of work (marketing), I decided to really impress Katr and sign up for an OpenAI account. Sort of a combo "Look, I'm trying new things!" and "keep your enemies closer" tactic, because Katr - and robots - respect that.
The first task I prompted ChatGPT to do was to write me a multiple choice trivia question about Formula One driving.
It said "Sure!" and wrote me a multiple choice trivia question about which F1 driver had won the most World Championships. The answer was Lewis Hamilton.
I've heard of Lewis Hamilton so I thought "This seems right." But I knew the people answering this question would FLIP THEIR SHIT if the answer was wrong, so I decided to double check ChatGPT's work.
It turns out that Louis Hamilton has won 7 World Championships! But you know what? So has Michael Schumacher. They're tied. And there was zero mention of it in this trivia question.
So I went back to ChatGPT and said "Hey, can you please modify the question to indicate that in fact there are two people who are tied for most World Championship wins?"
ChatGPT said "Of course!" and then rewrote the exact same question and then in the the answer wrote "It's Louis Hamilton, who by the way is also tied with this other guy."
This is the point where I understood that I probably should not use ChatGPT for anything factual. You know what it's great at? Bad puns and emojis. The emojis!! It throws shit in there that that I did not even know existed! AI has taken my emoji game to the next level. ⛲🏟️🪐
Something else that ChatGPT is very good at is helping with simple tasks, like reformatting documents. So last week, Katr and I were working on a project where we were creating a list of Olympic athletes, along with a few fun facts about them, such as how many Olympic Games they had attended, their most recent notable result, and a personal fact, like they play the trombone.
We were doing this list in a spreadsheet. But when we had to get it translated into French, we had to reformat it from a spreadsheet into Word, because Translation will not work from a spreadsheet.
So rather than ask one of our overtaxed team members to take on this tedious task, Katr asked ChatGPT to do it.
The desired format was very straightforward:
Name of athlete
Number of games attended
Notable result recent
Fun personal fact
ChatGPT made fast work of this. AI FTW! We did a quick scan of the resulting document to make sure it was formatted correctly and sent it off to Translation.
When the document came back from Translation, I had to spend a lot of time painstakingly editing down the French, because our pithy English "Santiago 2023: Silver in men's team foil" got translated as "Santiago 2023 : il a remporté l'argent à l'épreuve masculine de fleuret par équipes". So it took me a day to get to page 2 of the document.
I was using the English from the spreadsheet and the French from Translation. The next athlete on my spreadsheet was named François. Let’s say his last name was Cotton. Next on the list in the French document was the name François Lamontagne.
This was confusing. But I thought "Hey, there are a lot of moving parts in this project. Maybe this guy changed his name and Katr updated it in this document but not in the original spreadsheet?"
So I continued with François, comparing the French to the English version in the spreadsheet.
EN: First Olympic Games
FR: First Olympic Games.
Check.
EN: Currently ranked fifth in the world
FR: Canadian team’s top scoring player in mixed sabre.
WAIT.
EN: B.A. in Engineering.
FR: Previously competed in figure skating.
WHAT.
THE.
FUCK.
I continued reading down to the next athlete on the list. The name was unfamiliar. I looked back at our spreadsheet. And found the name did not exist there.
Neither did the next 11 names in the Translation document.
With growing horror, I started to grok what had happened.
Somehow, in the midst of reformatting a spreadsheet to a Word doc, ChatGPT had created 11 COMPLETELY FAKE athletes, complete with Games information, competition information and personal facts and, in the process, had DELETED SIX ACTUAL ATHLETES and their information from the list.
Here, for your reading pleasure, are some of the fake athletes ChatGPT created instead of just moving our actual athletes from a spreadsheet to Word:
Gina Daniels
First Olympic games
Represented Canada at 2019 Pan Am Games
Hobbies: photography, reading
Ivan Kojevnikov
First Olympic games
Won Bronze at 2023 European Championships
Studying computer science at UBC
Janet Lin
First Olympic games
Scored the winning goal at 2023 World Cup
Dream job: astronaut
Jesse Holman
First Olympic games
Named MVP at 2023 Nationals
Also plays the piano
Jessica Rodriguez
First Olympic games
Double Silver medallist at 2023 Pan Am Games
Speaks three languages
John Barnes
First Olympic games
Bronze at 2023 World Athletics Championships
Holds national record for javelin throw
Jordan Smith
First Olympic games
Won Gold at 2023 Nationals
Nickname: Jordo
I’ve obviously heard about ChatGPT just making shit up. The most well-known story is probably about that lawyer who got ChatGPT to write a brief and ChatGPT made up a bunch of law cases that didn’t exist. And obviously my experience with trivia question it made me leery of asking it to do any kind of factual research.
But reformatting a spreadsheet into a Word doc doesn't seem like the kind of request that would cause ChatGPT to "freestyle". So I’m sharing this story because it’s both funny and terrifying and because I know some of you out there like to use ChatGPT for these kinds of simple tasks and you assume that it's not going to fuck you. But it WILL.
So I'm taking a stand. We can’t trust the machines. We’re going to have to keep doing our own fucking jobs, guys. And, as our colleague Brhu says when we're trying to figure out if an image looks AI generated: "Check the fingers."
And now, a 50x50 update:
Dips in the pool: 16
French lessons: 25
Books read: 11
Air fryer uses: 12
Pillows touched: 45 (almost there!)
