Researchers have developed a new technique called ZeRO, which uses DeepSpeed and FairScale to improve the efficiency of training large AI models. ZeRO stands for 'Zeroing out' and it achieves this by reducing memory usage during training. According to its creators, ZeRO can train larger models in less time than traditional methods. --- Why it matters: This matters because it enables researchers to train more complex AI models faster and with less computational resources, which is crucial for advancing the field of deep learning. Source: https://huggingface.co/blog/zero-deepspeed-fairscale