Why Chatbots Always Pick the Number 7

Imagine walking into a room full of people and asking them to pick a random number between one and ten. You would naturally expect a chaotic mix of threes, fives, eights, and nines. But if you pose that exact same question to the world's most advanced AI chatbots—whether it is Claude, ChatGPT, or Gemini—they will almost unanimously shout "7".

This uncanny alignment is not a coincidence. It is a perfect illustration of a structural quirk in Large Language Models (LLMs) known as "groupthink." While we often treat these systems as boundless engines of imagination, the reality is that they are highly constrained by their underlying architecture. Because models are optimized to generate the most statistically probable text based on vast troves of human data, they naturally gravitate toward the most average, predictable responses.

Why does this matter? It comes down to what we are actually using AI for. If you are debugging a complex Python script or summarizing a dense financial report, predictability is exactly what you want. You need the model to be reliable and precise. However, this statistical safety net becomes a massive liability for creative work. If you are using AI to brainstorm a new marketing campaign, write a novel, or plan an unconventional European vacation, standard LLMs will likely feed you the safest, most cliché options available. They are programmed to be helpful, but in doing so, they become painfully unoriginal.

An Australian startup named Springboards is actively tackling this homogenization of thought. Recognizing that the market is saturated with models trained to find the single "best" answer, they have developed an AI model called Flint. Flint is engineered specifically to resist the gravitational pull of the obvious.

When presented with open-ended prompts—like asking for recommendations on where to travel in Europe—Flint is trained to generate a wider, more eccentric variety of answers rather than defaulting to the statistical mean of Paris or Rome. It pushes back against the standard algorithms by prioritizing cognitive diversity over sheer probability.

As artificial intelligence increasingly becomes our default brainstorming partner, the risk of standardizing human creativity is a genuine concern. If every writer, marketer, and planner relies on the same algorithmic consensus, our collective output will slowly lose its unique edge. The next frontier in AI development isn't just about making models smarter, faster, or more accurate—it is about teaching them how to think outside the algorithmic box and reclaim the unpredictability that makes true innovation possible.

Key Points

• Leading AI models exhibit 'groupthink,' consistently offering highly predictable answers to open-ended questions.
• This predictability stems from AI training methods that favor statistical probability over creative variance.
• While useful for technical tasks, AI groupthink severely limits its utility as a brainstorming or creative tool.
• Australian startup Springboards built 'Flint,' an LLM designed specifically to generate diverse and non-obvious responses.

Why It Matters

The homogenization of AI responses threatens to limit human creativity as we increasingly rely on these tools for brainstorming. Overcoming AI groupthink is essential for unlocking its true potential as an ideation partner.

Sources:

• The Download: a startup has a solution for AI’s groupthink problem — MIT Technology Review - AI