Why Most AI Chatbots Stay Predictable and How One Startup Plans to Break the Mold

Ask any AI chatbot for a random number between 1 and 10, and you'll almost certainly get '7'. This is no coincidence. It's a reflection of a fundamental tendency in large language models (LLMs) to stick to familiar patterns. Predictability is the name of the game, and it's a game these models seem unable to break out of.

Unmasking the Repetitive Nature of AI

In an experiment that's captured attention, Alex Bingemann, cofounder of the Australian startup Springboards, demonstrated this predictability by asking popular AI models like ChatGPT and Claude for random numbers. The usual result was 7. But when Bingemann put Flint, their own creation, to the test, Flint broke the cycle by offering up 3.7916 instead. This wasn't just a fluke. It highlighted a deeper problem: why, despite their sophistication, do so many LLMs converge on the same cliched answers?

Flint is an LLM designed with intentional randomness at key points. It's built on Qwen 3, an open-source model from Alibaba, and has garnered attention for its ability to provide more diverse responses without sacrificing coherence. Bingemann quipped, "We welcome hallucinations," admitting Flint's tendency to inject randomness offers a refreshing alternative.

This pattern isn't limited to numbers. When challenged to come up with metaphors for time, over half of more than 70 LLMs independently chose "Time is a river." The lack of diversity can be traced to the way these models are trained: similar data, tasks, and objectives lead to a hive-mind effect.

Implications for Creativity in AI

So, why does this matter for the world of crypto and beyond? Predictability in AI could be stifling innovation where it's most needed. In industries like finance or marketing, fresh ideas fuel growth and differentiation. With models like Flint, creative professionals might find new inspiration, avoiding the stagnation that comes with overused tropes.

Imagine a crypto project seeking a fresh marketing angle. Standard AI suggestions may lead to recycled themes. But with a model trained to surprise, there's potential for groundbreaking campaigns that stand out. It's not just about having a different output. it's about how these results can stimulate human creativity, pushing boundaries and challenging norms.

Here's where Flint's approach to randomness at specific decision points, as opposed to across-the-board chaos, shows promise. This method avoids incoherence while still adding much-needed variety. Can this approach spark an evolution in AI-assisted brainstorming? It might just be the catalyst the creative sectors need.

The Path Forward for AI Diversity

The question remains: will the rest of the AI world take note? If LLMs can shift from predictable to potential-rich tools, the users, particularly in dynamic fields like crypto, marketing, and finance, could find themselves as winners. However, there's a cautionary note here. Relying too heavily on AI, even one as fresh as Flint, can lead to complacency. As Maximilian Weigl from Uncommon cautions, "Most people are fine with good enough." The challenge is encouraging users to push beyond "good enough."

The takeaway? While Flint represents a promising stride towards more diverse, inspiring AI outputs, it also invites a broader consideration of how we integrate AI into our creative processes. Are we ready to embrace the oddball, the unexpected, and the unfamiliar? If creativity is king, perhaps it's time the crown slipped just a little.