Large language models (LLMs) are predictably predictable. A simple request for a random number or car brand often yields the same answers, making brainstorming tedious.
Australian startup Springboards has developed Flint, an LLM trained to offer diverse responses, aiming to break the monotony of groupthink. When tested, Flint delivered unexpected results like a Ford F-150 instead of a Toyota or Honda, and a unique tagline for New Balance running shoes.
Research highlights this issue, showing a high degree of repetition in LLM answers across different models. This may stem from similar training methods and data sets, leading to homogenous outputs despite varied inputs.
To combat this, Springboards offers a tool that combines text from multiple LLMs like ChatGPT and Claude, allowing creative professionals to explore diverse ideas more freely. Early feedback suggests Flint could be a game-changer for those seeking variety in their responses.
