The Unpredictabilities of Transfer Learning

There's a lot of work that can be done to even improve existing data sets, right? Even what was considered a very high quality dataset, three years ago, like the pile, then somebody else looks at it and figures out that there're a lot of ways to improve it. The hardest place to help is compute because you have to be in the hardware space,. Maybe the easiest place for people to help is in data.