LLMs are stuck in a groupthink groove. This startup is trying to get them out.

AI chatbots like ChatGPT and Claude have a hidden bias toward giving the same answers to repetitive requests—they pick numbers like 7 or 3 far more often than truly random distribution would suggest. A startup is working to fix this quirk, which matters because it reveals limitations in how 'intelligent' these systems really are.