LiGO is a new machine-learning technique developed by MIT researchers that cuts by about 50 percent the computational cost required to train large vision and language models.
