ai research feels experimental, like artisans in edo japan refining a craft over generations, or how economic theories emerged after the fact to explain what was already happening. trying to estimate how compute and cognitive labor substitute, and how that impacts timelines for asi. fascinating to see these patterns across different fields, from economics to tech to observing how kids gain independence walking to school. sometimes an idea clicks, connecting mokyr's knowledge systems to current ai trends, or a personal memory brings surprising clarity about control and learning. there's excitement in sharing tentative conclusions, even if they're rough around the edges or based on a hunch. piecing things together from research trips, late-night model tweaking, and paying attention to the world.