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

The piece illustrates how large language models often produce similar answers, even for simple tasks like generating a random number, highlighting a groupthink bias. It outlines a startup’s effort to increase response diversity in LLMs through novel techniques.

Summary written by Kernelia from the original article by MIT Technology Review. The story and its rights belong to its author.