A Data Set Problem and a Social Problem in Machine Learning Research?

I think there are a lot of the ingredients are there, but I haven't seen anyone really put them together in an interesting way. It's always nicer if you had actually a data set that captured all of these different patterns and it could learn it in an end to end way. So we can actually see the entirety of academia and research institutions as one algorithm that's performing lifelong learning by generating better and better architectures and better data sets. And over time, we'll solve the problem of AGI.