Most chatbots hand you the same safe answer everyone else gets. A startup called Springboards thinks that’s a real problem, and it built a model called Flint to fix it. According to MIT Tech Review, Flint is designed to break large language models out of the groupthink groove that pulls every answer toward the average.
Here’s what makes Flint different. Springboards figured out that the usual knob for adding creativity, cranking up the model’s randomness, is a blunt tool. Turn it up across the board and the whole response gets messy. What you actually want is more variety at a few key moments, not in every word.
🎯 How it works
Springboards took Qwen 3, an open model, and trained its own version to spot the exact points in a response where more variety makes sense. Then it fills those spots with words or phrases that are a little more unexpected.
The team gives a simple example. Ask a chatbot “Where should I go in Europe?” and the model only needs to loosen up right before it names a destination. Everything else in the sentence can stay normal. Flint targets that one moment instead of scrambling the entire answer.
“Flint’s programmed to throw an oddball in. It’s more of an invitation to think wider,” says Maximilian Weigl, cofounder and chief strategy officer at the marketing firm Uncommon, quoted by MIT Tech Review. His team runs Flint next to ChatGPT, Claude, and Gemini. “You can’t really create something boundary-breaking with tools that pull you back to the average,” he says.
🧭 Who it’s for
Right now Flint points squarely at advertisers and marketers, because those are Springboards’s customers. That’s a smart place to start. Ad and creative teams live or die on ideas that don’t sound like everyone else’s, so a model built to spike variety on demand fits their work.
But the founders, Bingemann and Browne, argue the boring-average problem hits anyone who uses chatbots, not just marketers. Their pitch is about giving people the choice. “Variety is great when you’re trying to spark ideas,” Bingemann told MIT Tech Review. “Let’s go down this route instead of letting the machines do it all and ending up in a gray, boring world.”
⚠️ The honest caveats
What stands out here is that even a happy customer waves a caution flag. Weigl notes that nine times out of ten, the average answer is fine. “Most people are fine with good enough. They want to see mass-market familiar things,” he says. You don’t reach for extremes on every task, and a tool tuned for oddballs isn’t always the right call.
He goes further on the risk of leaning on any AI too hard. “I have a big problem when people rely on the output from any AI, including Flint,” he says. “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.'”
That’s a useful reality check coming from someone selling the value of the tool, not against it.
💡 Why it matters
Flint is a bet on a real weakness in today’s models. As more people prompt the same handful of chatbots, outputs converge. Everyone gets nudged toward the statistical middle, and content across the web starts to blur together. A model that adds controlled unpredictability, only where it counts, is a genuinely different approach than just sliding the temperature up.
The targeted method is the interesting part. Plenty of tools let you dial randomness. Training a model to decide where variety belongs is a more surgical idea, and it’s the kind of thing that could matter well beyond ad copy if it works.
The open question is whether “controlled surprise” holds up outside marketing, where a wrong-but-interesting answer is a feature. In fields where accuracy beats novelty, throwing an oddball in is exactly what you don’t want. For now, Flint is a focused product solving a focused problem, and it’s worth watching whether Springboards can widen it without losing the point.
More details are available in the original MIT Tech Review report.