ChatGPT loses at pun competition
This one is for you Nir
BROOKLYN—I heard the MC call my name and felt my legs carry me toward the stage. It was time to enter the Punderdome.
I’d never competed in a pun contest, much less in front of hundreds of people at an event considered the Roman Colosseum of punditry. My stage presence could be described as lacking. I had done basically no preparation. I did, however, have one thing going for me: I was actually a robot.
image Or, rather, its assistant.
ChatGPT, the trendy new artificial intelligence robot, had generated all of my puns. It’s a crazy good chatbot. So good, in fact, that it has some folks calling this the end of the human race as we know it.
The chatbot can write an essay on Proust in seconds. Want a limerick about the Cold War? It can rhyme “tensions ran high” with “nuclear sky.” In one widely spread example, it dished out instructions on how to get a peanut butter sandwich out of a VCR, written in the style of the King James Bible.
Could it match the wit of a human pun champion? I was about to find out at Punderdome, a regular pun contest that draws big crowds to a performance venue in Brooklyn.
A skillful pun competition between two people sounds more like a conversation with a heaping dose of puns about a topic slipped in. In one YouTube video I watched the night before the event, two punners faced off on the topic of dog breeds. “I found that some instruments you can carry with you everywhere. But a bass? Set it down,” one said (basset, get it?). The other shot back: “Does that bass play a sharp A?” (Shar Pei, obviously).
I asked the chatbot for help. “Tell me a pun,” I typed in. “Why was the math book sad? Because it had too many problems,” it answered. More of a dad joke than a pun, I thought. It was the first of many times the bot would spit out that answer.
My colleagues and I typed in different prompts, but struggled to get anything particularly witty. “Word play about Kalamazoo” returned snores such as “Kalamazoo: where history comes alive.”
How do the real punsters do it? Before Allison Fisher started competing at Punderdome under the name Rhyme & Punishment five years ago, she went to a coffee shop with a friend. They went back and forth practicing two-minute monologues the way they’re done in the show. She won three times.
“It’s really all about noodling around the ideas in your head,” said Ms. Fisher, who is a software engineer. “After thinking for 15 seconds orzo, I’ll take a penne to paper. I’ll come up with a fu-silli ones.”
Emma Taylor Miller, who has a degree in drama and does some side work as an actor and clown, met her boyfriend when he introduced himself with a joke through an online dating website. “Did you hear about the explosion at the French cheese factory? There was de-Brie everywhere.” Her response: “That’s a Gouda one.”
During the week before she competes under the stage name “When Wit Hits the Fan,” she plays a Punderdome card game that contains prompts to get the mind punning.
Watching videos of pun-offs, it was clear that one key to designing a pun that would land was to start with the punny word and work your way back toward the setup.
Would the robot know that? I had a little hope. Watson, the supercomputer built by International Business Machines Corp., managed to beat “Jeopardy!” champions in 2011.
“It’s not trivia,” Erika Ettin, aka Lexi Kahn, corrected me while we were waiting for the show to start.
Fred Firestone co-founded the Punderdome in 2011. His daughter was a burgeoning comedian at the time and decided to run a pun competition, so she asked him for help. He flew in from St. Louis on a few days’ notice. He has been doing so almost every month or two since, even after his daughter went on to other endeavors. I would be in his 135th Punderdome.
When I called him on behalf of the robot, he was game for testing out its chops, so we designed an experiment. He sent my pun topic to my colleagues the afternoon before the show: cities and states. They asked ChatGPT to generate a bunch of puns and put them in a sealed envelope.
Mr. Firestone told the audience of 250 about this unusual plan, and made clear I wasn’t a ringer. My turn would be part of a one-off round separate from the night’s competition.
“Ben, just to be clear, brother,” he asked me on stage, “Are you a punner at all? You have any proclivities, any abilities in the punning arena, yes or no sir?”
“Absolutely not,” I replied.
The audience, apparently unthreatened by the robot overlords, let out some cheers. “Come on, Ben!” a few people shouted. “Wooh!”
A bit weak-kneed, I opened the envelope. I had 20 minutes during the intermission to read through the ChatGPT’s results and select the best puns. I wrote them on a mini whiteboard, which was my only allowable prop.
ChatGPT didn’t have much to offer. “In Peoria, the corn is so sweet it’s almost sinful.” Huh?
I wrote a few passable puns on the whiteboard along with some that were so bad that maybe they’d draw chuckles.
Next, I had to pick my competitor. Mr. Firestone invited up any past winners who wanted to participate. Quite a few rushed to the stage. “Any other champs want a piece of this guy?” Mr. Firestone asked.
He asked me to choose who I wanted to play against. I figured, if I was going to lose, I might as well lose to the best. I chose Nikolai Vanyo, a writer and filmmaker who onstage goes by Daft Pun. He was one of the top three biggest winners ever. “This is for all of you humans,” he told the crowd.
The spotlights were on us as we took position at twin mics. We would be going back and forth in a pun-off for two minutes. I held my breath.
“I’m not from the shore, but I Jersely know how to have a good time.” The crowd chuckled. Why? I don’t know. I was so nervous I transposed shore and Jersey.
Mr. Vanyo shot back: “I don’t like to not drink. I hate to Miss-is-sippi.” The crowd laughed louder.
I had that state on my whiteboard. “Oh, how I Mississi-thee,” I said. The robot was vaguely getting the hang of it.
Or was it? I found myself saying soon after: “New York City is the big apple. New Jersey is just another basket.” “What?” someone from the audience shouted. I was so embarrassed, I felt the need to add: “Chatbot speaking.”
Mr. Vanyo was picking up steam: “I was given the choice recently between a bag or a little mint—a sack-or-a-Mento.” (Say it again, slowly.)
I decided to use the robot’s best pun: “What’s the state where common sense is in short supply? Flori-duh.” The crowd loved it. I was enjoying myself. I can’t speak for the robot.
A few more back-and-forths and our allotted two minutes expired. Mr. Firestone asked if we wanted to go for another minute. I had used up everything remotely punable. But the crowd started cheering. So I consented.
“Go ahead, My-ami,” Mr. Vanyo said.
I tossed out a random one I had jotted down last minute even though it wasn’t actually a pun. “Boise, where the potatoes are always hot and the people are always friendly,” I said.
“I think between me and the robot, I-da-hoe here,” he said.
The robot never recovered.
Once the time ran out, a designated audience member came out, put on a blindfold and wore a “clap-o-meter” to judge which contestant got the most applause. The winner was obvious. I blamed my master, the robot, for giving me such thin material. The audience seemed sympathetic.
“You just work here!” someone shouted.
“I think I expected more from the bot,” Mr. Vanyo told me the next day. He said he had been punning so long that he had come to see the structure as mechanical, something a robot could replicate.
A spokeswoman for OpenAI, which created ChatGPT, pointed me to a blog post by a company employee that suggested a future in which creative endeavors could harness both the objectivity of AI and the personal narrative of humans.
Perhaps the robot’s assistant was the failure here.
As it turned out, I wasn’t the first one to try to get a computer to do my punning. Max Parke, a long time Punderdomer and software engineer, once tried to write a program that could get a computer to make puns. He gave up pretty quickly.
He said that the best puns are the most surprising ones and it’s hard for a computer to go off in different directions that it hasn’t seen before.
Ms. Miller said she thought maybe the computer didn’t recognize how much words and language can be mutated when spoken. Ms. Fisher said she thought maybe the computer would have done better if it was fed transcripts of past Punderdomes.
“Maybe a computer can server up some good puns,” Mr. Parke said. “But the ones I C? PU!” (Sorry, just to explain, a central processing unit is the brains of a computer.)