A new paper argues that the computing demands of deep learning are so great that progress on tasks like translation and self-driving is likely to slow.

Read More