If you ask different AI assistants to pick a number between one and ten, most will give you the same answer. It turns out that today’s most popular AI tools are not just smart—they are remarkably predictable. This tendency to gravitate toward popular or safe answers is now being called out as a form of digital groupthink.
The problem is that many AI models are designed to find the most probable answer to a prompt. This is efficient for tasks like writing code or summarizing a meeting, but it is lackluster when you need a creative idea or a unique travel recommendation. An Australian startup called Springboards is trying to break this habit. They have released a tool called Flint, which is specifically trained to reject the most obvious, statistically likely responses in favor of more varied, imaginative conclusions.
Giving AI a creative nudge
To understand why AI settles on the same answers, think of a model as an extremely well-read apprentice who has memorized a vast library of text. When you ask it a question, it doesn't actually think in the human sense. Instead, it looks at the sequence of words patterns it has learned and calculates the most likely next word, over and over. Because it wants to be helpful and avoid making mistakes, it heavily favors the most common patterns it saw during training. It is essentially choosing the safest, most statistically mainstream thing to say.
Springboards is addressing this by changing the internal rules, or parameters, that govern how the model picks its words. By instructing the model to prioritize variety over pure probability, they force it to veer off the well-trodden path. Think of it less like a librarian who always suggests the most popular bestseller, and more like an improv actor instructed to always avoid the first joke that comes to mind. By intentionally suppressing the most likely outputs, the system explores less common, but often more interesting, connections.
We often treat our chatbots like impartial experts, but the source material reminds us that they are products of their training. If an AI is only programmed to optimize for the safest, most middle-of-the-road answers, it can unintentionally narrow our own thinking. When we rely on these tools for planning or decision-making, we might be getting a consensus of the internet rather than a truly helpful or creative suggestion. As we bring these tools deeper into our lives, the ability to steer them toward nuance and novelty might become just as important as the raw information they provide.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy