Researchers have introduced a technique for compressing a large language model's reams of data, which could increase privacy, save energy and lower costs. The new algorithm works by trimming ...