Up to 30% of the Power Used to Train AI Is Wasted: Here's How to Fix It
University of MichiganA less wasteful way to train large language models, such as the GPT series, finishes in the same amount of time for up to 30% less energy, according to a new study from the University of Michigan.