LLMs are stuck in a groupthink groove. This startup is trying to get them out (www.technologyreview.com)

An Australian startup named Springboards has launched an innovative large language model (LLM) called Flint, designed to break away from the predictable, groupthink responses typical of mainstream LLMs like ChatGPT and Claude. Flint has been engineered to provide a wider variety of answers to open-ended questions, which could be particularly useful in creative fields such as advertising and marketing. By leveraging an open-source model called Qwen 3, Flint introduces randomness strategically at specific points in its responses, allowing for more diverse outputs without sacrificing coherence. This development is significant for the AI/ML community as it highlights a common limitation in current LLMs: their tendency to produce repetitive and similar answers due to similar training data and methodologies. A recent study titled "Artificial Hivemind" revealed that even when queried with diverse prompts, many models yielded strikingly similar responses, stifling creativity. By enabling users to generate novel ideas and consider different perspectives, Flint could help revitalize creativity in brainstorming sessions and project planning, ultimately leading to more unique and impactful outputs in various industries.