Australian startup Springboards has built an LLM called Flint to break the groupthink habit of mainstream chatbots. When users ask Claude, ChatGPT or Gemini for a random number, the answer is almost always 7, showing how predictable large language models have become. Flint is trained to generate a broader set of responses to open‑ended prompts such as travel recommendations, aiming to restore creativity in brainstorming sessions today.

Current LLMs excel at deterministic tasks like coding or research, but their conservatism hampers creative ideation. Flint’s architecture injects stochastic sampling and diversified training data, producing varied outputs without sacrificing factual accuracy. Early demos show the model suggesting unconventional European itineraries instead of the usual Paris‑Rome‑Barcelona loop, demonstrating tangible benefits for planners and marketers.

By forcing LLMs out of their safe‑zone responses, Springboards positions Flint as a tool for any application that values novelty, from product concepting to travel planning. The approach challenges the prevailing bias toward uniformity in AI outputs and offers developers a plug‑in that can be swapped into existing chatbot pipelines. Flint already powers a beta version of Springboards’ own brainstorming assistant.