LLMs Are Stuck In A Groupthink Groove This Startup Is Trying To Get Them Out

Large language models often produce nearly identical answers to the same prompts, creating an AI echo chamber that influences search results and medical advice alike.

A startup is fighting this by feeding models deliberately contradictory datasets, forcing them to learn multiple valid perspectives rather than converging on a single consensus view.

Early tests show these models gave noticeably different answers to complex questions and even flagged errors in each other's reasoning during collaborative tasks.

If we can teach AI systems to genuinely disagree without breaking down, could we finally build machines that reason like a panel of experts instead of an echo chamber?