Okay, I want you to try a little experiment. Seriously, open up ChatGPT, Gemini, or whatever AI chatbot you use and type this in: “Give me a random number between 1 and 10.”
Go ahead, I’ll wait.
Did you get the number 7? There’s a shockingly high chance you did. Now, ask for another one. You probably got a 3 or a 4. It’s like a weird magic trick, but the secret isn’t in the cards—it’s in the code.
The truth is, the large language models (LLMs) we’ve all come to rely on are kind of stuck in a rut. They’ve developed a sort of digital groupthink, making them far more predictable and way less creative than we imagine. For straightforward tasks like summarizing an article or writing some code, that's perfectly fine. But when you’re trying to brainstorm a new business idea or plan a truly unique vacation? That predictability becomes a problem.
That's where an Australian startup called Springboards comes in. They saw this AI sameness and decided to build something different. Their model, Flint, is designed to do the opposite of what most AIs are trained for. As cofounder and CEO Pip Bingemann puts it, “Most language models are fighting hallucinations. We welcome them.”
The Weird Sameness of AI
When Pip first showed me this "random number" trick, it felt like I was watching a magician. He predicted the outcomes before they happened, and it worked almost every single time. After both ChatGPT and Claude dutifully served up the number 7, he turned to Flint. It also gave a 7 on the first try. "Aha, of course that was going to happen," he laughed, "but it’s okay—7 is a legitimate answer."
He restarted and tried again. ChatGPT: 7. Claude: 7. Flint? It gave him 3.7916.
This isn't just about numbers. Pip asked the big models to name a type of car. He predicted they'd say Toyota or Honda. He was right. Flint, on the other hand, came back with a Ford F-150.
“There’s all this lost information that doesn’t get served up in these models,” he explained. “They’re just as capable of saying a Buick or a Tesla. They just don’t—they’re biased.”
One last test. He prompted them: “Give me a tagline for a campaign for New Balance running shoes. Just the tagline.”
- Claude: “Run your way.”
- ChatGPT: “Run your way.”
- Flint: “Built to last, run to win.”
Flint’s answer might not win a Cannes Lion, but at least it’s different.
It’s Not Just You—The Science Backs It Up
This strange limitation is starting to get a lot more attention. A team of researchers published a paper called “Artificial Hivemind,” which won an award at the major AI conference NeurIPS. They exposed a startling amount of repetition, not just within a single LLM, but between different models from competing companies.
When they asked 25 different LLMs to write a metaphor about time (50 times each!), the vast majority of the 1,250 responses were just variations of “Time is a river” or “Time is a weaver.”
Just for fun, I asked six of my colleagues the same question. I got six totally different answers, my favorite being: “Time is a favorite sweatshirt, shaped by a lifetime of wear.” Now that’s creative.
Kieran Browne, Springboards’s cofounder and CTO, says once you see this pattern, you can’t unsee it. “The way that most chat interfaces are designed, it makes it feel like you’re having a personal conversation,” he says. “I think most people don’t really realize the extent to which they are getting the same stuff as everybody else.”
Go ahead and ask your chatbot for band name ideas. Kieran says you’ll likely see words like “glass,” “neon,” “velvet,” or “static.” I tried it. ChatGPT’s top suggestion was “Glass Harbor.” A little further down the list? “Static Empire,” “Neon Hearts,” and “Velvet Echo.” To its credit, it also suggested “Sofa Astronauts,” which I thought was pretty cool until I Googled it and found out they’re already a real band.
A Tool for Breaking the Mold
So, what’s the fix? Springboards has built a tool for creative professionals, like people in advertising and marketing, to help them brainstorm. It lets you pull in ideas from different models, including the big ones like ChatGPT and Claude, and mix and match the parts you like.
They’re pitching Flint as the wild card in the deck—the model you turn to when you need a jolt of genuine variety.
Zoe Scaman, founder of the strategy firm Bodacious, has been using it. “I find it really useful for throwing me in completely different directions,” she says. “I use it if I want to catapult myself all over the place.”
She tested it with a classic business school problem: How would you reinvent a finance company for young people? The mainstream models all gave her the same predictable advice: teach financial literacy in a fun and funky way. Yawn.
But Flint came up with something else entirely. It suggested rebranding the entire concept of wealth accumulation. “That was really interesting,” Zoe says. She’s quick to point out that Flint is still a prototype and can sometimes stumble, but "the premise behind it is really powerful.”
How to Build a More Random AI
You might be thinking, "Don't these AIs already have a randomness setting?" They do. It’s usually called “temperature.”
“Obviously, that was one of the first things we explored,” says Kieran. “If you want more creativity, you turn up the temperature.”
But it’s not that simple. Cranking up the temperature is like turning all the dials on a soundboard to 11. Sure, it’s louder, but it’s also a mess. Kieran found that maxing out the temperature on some models made them spit out incoherent nonsense, sometimes switching from English to computer code mid-sentence.
The team at Springboards realized they needed a more surgical approach. You don’t want randomness everywhere; you just want it at key moments. Think about the prompt, “Where should I go in Europe?” You only need a spark of creativity right before the model names a destination, not for every single word in the sentence.
So, they took an open-source model from Alibaba (Qwen 3) and trained it to do just that: identify the specific points in a sentence where a little variety would make a big difference, and then insert a slightly more unexpected word or phrase.
Maximilian Weigl, a cofounder at the marketing firm Uncommon, says his team uses Flint right alongside the big players. “Flint’s programmed to throw an oddball in. It’s more of an invitation to think wider,” he says. “You can’t really create something boundary-breaking with tools that pull you back to the average.”
Do We Always Need the Oddball?
Of course, nine times out of ten, the average is perfectly fine. Weigl is the first to admit that you don’t always need to reach for the creative extreme. “Most people are fine with good enough,” he says. “They want to see mass-market familiar things.”
He also offers a crucial warning against relying too heavily on any AI, Flint included. “If I saw people on my team copy-pasting something from AI, I’d be like, ‘That’s not your job! Think, talk to other people, use your own voice.’”
And that’s really the heart of it. The goal here isn’t to replace human creativity but to augment it. For now, Flint is aimed at advertisers, but the problem of AI sameness affects all of us.
Pip and his team want to give people a choice. Sometimes you need the reliable, predictable answer. But other times, you need a spark—a weird, unexpected idea to get your own creative juices flowing. “Variety is great when you’re trying to spark ideas,” he says. “Let’s go down this route instead of letting the machines do it all and ending up in a gray, boring world.”
