Google’s new AI trick cuts hallucinations without killing usefulness
Google researchers have developed a method that lets large language models admit uncertainty without abandoning helpfulness. The approach could bridge the gap between trustworthy AI and practical deployment.