Characterizing Higgs pair production has important implications for our understanding of the Higgs potential and the early universe. We discuss new techniques for identifying highly Lorentz-boosted Higgs bosons with the CMS experiment at the CERN Large Hadron Collider (LHC), highlighting in particular novel transformer-based machine learning (ML) developments in jet tagging. We then demonstrate their impact in a new CMS result constraining the quartic Higgs-to-vector boson (HHVV) coupling in the all-hadronic two beauty quark (2b) and two vector boson (VV) final state. Finally, we look ahead to AI advances in simulating high energy collisions, using physics-informed and equivariant graph neural networks, and in the real-time CMS trigger to ultimately boost our sensitivity to such measurements and beyond in Run 3 and the high-luminosity era of the LHC.