Your model can absolutely improve
How would that work out barring a complete retraining or human in the loop evals?
How would that work out barring a complete retraining or human in the loop evals?