Why AI safety is broken, faith in large models is waning, and what it means to be “sustainably relevant.”