Today was a 21-cm cosmology meeting at Flatiron. Unfortunately I could only do the morning. Ue-Li Pen (CITA) spoke about reconstruction using a simplified dynamics. I suggested that anything that could be done with simplified dynamics could be done better with machine learning. I think I can even prove this, since the machine learning could be trained on the residuals away from the simplified dynamics reconstruction! In his talk, however, he mentioned this incredible Wang et al paper that does full reconstruction of the initial conditions for the entire SDSS Main Sample volume! This gives me hope for the future of cosmology.