Summary

Diego Rodriguez discusses the challenges of AI evaluation, focusing on human perception and aesthetic judgment in generative media technologies. He highlights a key example of an AI struggling to provide meaningful commentary on an obviously unnatural AI-generated hand image, revealing significant limitations in current AI systems' understanding. The talk explores how AI models are trained on human data and preference, yet often fail to capture nuanced human perception and intuitive responses. Rodriguez suggests the need for more critical questioning and evaluation methods in AI development that go beyond current technical metrics and incorporate deeper insights into human experience and perception.