Startup aims to cure LLMs of groupthink with diversity-boosting technique

A new startup is tackling the tendency of large language models to produce predictable outputs, like always picking 7 as a random number. The company's approach introduces methods to increase response diversity, potentially making AI more creative and useful. Early tests show significant reduction in repetitive patterns across major models.