Abstract: For centuries, our inability to read hieroglyphs condemn us to miss most of the information about three thousand years of history of the ancient Egypt. Today, most of the information contained in cosmological surveys about fundamental physics is lost because we lack the tools to read it. In this talk I will first show the wealth of cosmological information that resides on small scales, information that cannot be extracted using standard cosmological statistics. I will then discuss the challenges in extracting information in this regime, foremost non-linear gravitational dynamics and baryonic effects. Next, I will show how neural networks can learn optimal estimators to extract cosmological information while marginalizing over uncertain baryonic effects. Finally, I will describe how we are combining thousands of N-body and state-of-the-art magneto-hydrodynamic simulations from the Quijote and CAMELS suite to build a cosmological “rosetta stone”, that like with hieroglyphs, may allow us to decipher new information from upcoming astronomical surveys.