Startup launches tool to combat groupthink in large language models

The weekday newsletter The Download highlights a growing concern among artificial‑intelligence developers: large language models (LLMs) are increasingly producing homogeneous responses, a phenomenon described as “groupthink.” The briefing notes that the pattern emerges across major chatbots—including Claude, ChatGPT and Gemini—when users request simple, open‑ended tasks such as generating a random number. The uniformity of answers has prompted industry observers to call for mechanisms that can inject genuine variability into model outputs.

In response, a newly launched startup has introduced a tool designed to break the conformity cycle. By prompting users to ask any chatbot for a random number between 1 and a chosen limit, the service forces the underlying model to draw from a broader distribution of possible answers, thereby encouraging more diverse reasoning paths. The initiative aims to demonstrate that modest changes in user interaction can help LLMs escape entrenched response patterns, offering a practical step toward more varied and resilient conversational AI.