I spent the day at the Yale Center for Astronomy and Astrophysics (hosted by Tabby Boyajian), where I gave a seminar on data-driven models and The Cannon. I spent quite a bit of time distinguishing the model we use from vanilla machine learning; I really think that machine learning in its basic standard form is not that useful in astronomy (of course this depends in detail on your definition of the "scope" of "machine learning"). After my talk I did the rounds, and had many interesting discussions. One of these was with Andrew Hearin (Yale) who is working on generalizing the halo occupation model for large-scale structure modeling. He is convinced that we can't beat ten-percent-ish inferences about cosmology at 10-Mpc-ish scales unless we make the halo model far more general. At dinner, we had great conversations about the future of data-driven modeling and astrophysics. Thanks, Yale!