I've written a blog post about how scaling, as a geometric operation, can help us understand and analyze shapes and solve geometric optimization problems, e.g., interpenetration resolution. It brings together our work from Eurographics 2022, SGP 2024, and SIGGRAPH Asia 2026: frank-zy-dou.github.io/blog/… - Enlarge, then shrink: dilate geometric primitives to select a compact subset, then restore their original sizes to obtain a skeletal representation capturing the shape's main structural features. - Shrink, then enlarge: shrink objects to create clearance, then adjust their poses as they return to full size, resolving interpenetration along the way. Eight years ago, in 2018, I started doing research on geometric optimization. Back then, there was no ChatGPT or Claude to turn to. Junior researchers often had to spend considerable time setting up libraries such as CGAL, configuring optimization solvers, and reimplementing baselines themselves in C++. And if you decided to implement an idea in CUDA, that usually meant another substantial amount of engineering effort. So much has changed since then, especially in AI and in how we do research. I've followed many discussions about what these changes mean for researchers. Whatever our views, this shift is already here. Times have changed. Still, I hope we can hold on to "the dignity of thinking for ourselves." To me, this is really, really important. While we continue figuring out what research should look like in this new era, I'd like to set aside one AI-free day each week for some old-fashioned, independent thinking.