We used ai to make dogs play basketball. Here's how that went.

April 13, 2023 - 6 min read
ManyPest Madness

Well, the ManyPets Madness basketball tournament is in the books. 

16 baller breeds faced off in a single-elimination barker royale, and a surprise champion — the Powerhouse Pugs — emerged as top dogs.

To help generate game outcomes and recaps, we enlisted the AI tool ChatGPT. But making those game recaps funny, energetic, and surprising — well, that took some good-old-fashioned human creativity. In the end, we learned a lot about how AI can help with content creation, and when it needs an assist. 

And hey, maybe some of the lessons we learned can be useful to any other intrepid content creators attempting to bend AI to their will…especially if they’re creating a fictional dog-themed basketball tournament. 

Here are our top five tips for using AI to tell ridiculous stories.

1. AI Can’t Handle Succinct Humor and Wordplay  

When we started building out the ManyPets Madness tournament, first we decided to give each breed a cute team name. But as it turns out, AI isn’t all that great at snappy jokes and titles.  For example:

- ManyPets: Come up with a funny name for a basketball team made of Chihuahuas.

- AI: “How about “The Chihuahua Dunkaroos”?

Um, what? Like the discontinued snack food with the kangaroo on the box? Weird! And how does that name reflect anything about Chihuahuas, specifically? Why not the Alaskan Malamute Dunkaroos or the Three-Toed Sloth Dunkaroos? 

In the end, we went with the “Diminutive Dribblers,” a name we dreamed up with precisely zero input from Skynet ChatGPT. In fact, we generated all the team names, and the names of various fictional players and coaches (Lab-ron Jowls, anyone?), with our own squishy human thinky brains. 

2. It’s Okay to Set the Scene Yourself

Don’t expect AI to just magically know the backstory and mythology of your universe — you’ll have to help it fill in those details. If you simply ask ChatGPT to “tell a story where Pugs play Golden Retrievers in a basketball game,” the setting and overall results might be colorless and drained of energy. 

AI can only pull from existing sources. So if you’re dreaming up something entirely new — say, a ruff-and-tumble basketball tourney populated with colorful canines — that color has to spring from a human imagination. Here’s how we opened our ChatGPT prompt for the championship matchup: 

- It was the championship round of the ManyPets Madness basketball tournament. The Powerhouse Pugs had stubbornly willed their way to the finals. But they had a daunting obstacle ahead of them: The 27-time-defending-champion Golden Gamers, the Golden Retriever squad that hadn't lost a single game since President Bark Fetchton was in the Dog House.

Colorful scene-setting? Check. Fleshed out mythology and backstory? Yep. Wacky humor? Totally. AI began by repeating the first section of our prompt almost verbatim, then maintained its remarkably silly energy throughout the entire matchup.

3. Want AI to Be Funny? Tell It to Be Funny! 

Every prompt we gave to ChatGPT ended with the same three words: “Make it funny.” And you know what? The game recaps were super cute and funny! AI might not be great with snappy one-liners, but it’s not too shabby at telling funny stories. You just have to tell it what tone to use. Otherwise, it tends to recount events with a flat matter-of-factness that reminds you it’s just a robot.

It might sound ridiculous, but simply saying “make it funny” will prevent AI from treating a subject with the utmost seriousness. And when you’re talking about dogs playing basketball, serious is the last thing you want.

4. Feel Free to Counteract AI’s Strict Logic

ChatGPT did help us determine the winners of our matchups — but that doesn’t mean we simply let the AI go it alone. We wanted to keep our games fun and frisky, not just dig up a bunch of bow wow blowouts. 

One thing we learned: Left to its own digital devices, AI will probably take a literal-minded approach to picking winners, selecting the bigger, faster breeds against their athletically-challenged competitors. For instance, here’s what AI dribbled out in its first attempt to produce a Pug/German Shepherd face-off:  

- ManyPets: A team of German Shepherds plays a team of Pugs in a fictional basketball game. Who wins? Describe the game.

- AI: In this fictional game of basketball between a team of German Shepherds and a team of Pugs, it’s likely that the German Shepherds would win due to their size, athleticism, and agility. Pugs are much smaller and slower, which could put them at a disadvantage.

Ugh — stop taking everything so seriously, AI! 

On our second attempt, we added a new story variable — plus we gave it that tonal direction we discussed above — and that made all the difference.

- ManyPets: A team of German Shepherds plays a team of Pugs in a fictional basketball game. The German Shepherds find the Pugs’ sad eyes distracting. Who wins? Describe the game. Make it funny. 

Those tweaks changed everything, sending the game — the entire tournament, in fact — on a different trajectory. The Pugs emerged victorious, and a historic Cinderella story was set in motion.

(Our apologies to German Shepherd lovers.)

5. When It Comes to Word Count, Go Long — Then Short

Did you know you can give ChatGPT a word count? Super helpful. For design reasons, we wanted to make sure all our game recaps were under 250 words.

Here's the thing, though: AI is in dire need of an editor. When attempting to tell a rousing story, it often meanders and beats the same story points into the ground. Here’s a section from AI’s first draft of our Shiba Inu/French Bulldog matchup: 

- The game started off with the Shibas dominating, using their size and speed to get ahead on the scoreboard. Shiba Inus aren’t all that big, but they’re still bigger, faster dogs than French Bulldogs, and the Frenchies were struggling to keep up. Then everything changed when one of them perked their ears and suddenly seemed to know exactly where the ball was going. 

It made no sense. The crowd was confused at first, but soon realized what was happening. The  Frenchies were using their amazing hearing to anticipate the Shibas' moves on the court almost as if they were psychic. It was like they had some sort of superpower, and the Shibas were completely caught off guard.

The Frenchies were making plays that seemed impossible, stealing the ball from the Shibas and making shots that defied belief. The Shibas tried to fight back, but the Frenchies had them beat at every turn. It was as if their pointy little ears had granted them the powers of precognition, and there was simply no stopping them.

Funny — but a bit bloated and repetitive. We were ultimately able to cut this section in half without losing any substance:

- At first, the Shibas dominated. The Frenchies huffed and puffed, just trying to keep up. Then suddenly, one of them perked up his ears and seemed to know exactly where the ball was going.

The crowd was confused at first, but soon realized that the Frenchies were using their amazing hearing to anticipate the Shibas' moves on the court. The Frenchies were making plays that seemed impossible, stealing the ball from the Shibas and making shots that defied belief. It was as if their bat-like ears had granted them the powers of precognition.

Voila: 174 words became 93 words, giving us a lot more space to finish the game recap without exceeding 250 words. 

From that point forward, we started telling ChatGPT to aim for about 500 words instead of 250. Why? Because we’d learned that while AI can unearth some gems, it likes to be long-winded and inevitably needs to be edited down. If we’d started at 250 words and trimmed the fat from there, we would have ended up with some very short stories. 

Look, here’s the bottom line: AI can give you a decent-ish first draft. But it’ll need a lot of editing, and you’ll need to give it some heavy directions. AI isn’t going to take over content-creation, let alone the world, just yet. Which is great news for all of us here at ManyPets, because we’re already hard at work developing the 2024 ManyPets Madness tournament. 

David Teich
Lead Editor

David oversees content strategy and development at ManyPets. As Lead Editor, he focuses on delivering accurate information related to pet care and insurance. David’s editorial background spans more than a decade, including a pivotal role at Digiday, where he wrote content and managed relationships with media and tech companies. As an Associate Editor at Cynopsis Media, David wrote the Cynopsis Digital newsletter and interviewed executives and digital marketing experts in the TV industry. His background also includes film journalism. His diverse experiences in journalism and marketing underpins his role in shaping content within the pet care industry.