Weak lensing of photons by large scale structure is sensitive to both the growth of the (lensing) structures and the expansion history of the universe. The large number of modes available on small scales have the potential to constrain cosmological parameters beyond what is achievable in the linear regime, but extracting this information from weak lensing surveys is not straightforward. I will review how to do so emphasizing the use of non-Gaussian observables and discuss how deep convolutional neural networks offer a viable alternative. The amount of information is limited by our ability to model the matter density field on small scales, including baryonic effects. I will end by explaining how future high resolution CMB experiments can inform simulations by measuring the angular momentum of baryons in the circumgalactic medium.