Timing Trick Cuts Energy Used in LLM Training by Up to 14 Percent

# Training AI Models Just Got More Energy-Efficient Researchers have discovered a simple way to cut the energy needed to train large AI models like ChatGPT by up to 14 percent—without slowing down the process. By adjusting the speed at which computer chips (GPUs) process information during training, they found that many chips were running faster than necessary for certain tasks, wasting electricity in the process. This discovery matters because training advanced AI models currently consumes enormous amounts of power, and even modest improvements could significantly reduce both costs and environmental impact.