The real challenge is real world deployment. Great models may fail due to population differences, fairness issues bc training data is not representative. We need high quality data but we also need adoption and that depends on how you integrate solutions into existing work flows.