You are discussing model design choices for an AI system and want to explain how you think about training objectives.
What is the loss function you favor most in AI, and why?