According to Ethan Mollick on X, a widely cited Apple-affiliated paper from June 2025 that questioned AI reliability triggered significant debate but has proven less relevant over the last year as frontier models improved (source: Ethan Mollick on X). As reported by Mollick, recurring interest in so-called AI must fail or model collapse papers outpaces attention to studies showing strong model performance, reflecting industry discomfort with AI risks (source: Ethan Mollick on X). According to public discussion summarized by Mollick, the business takeaway is to benchmark current model generations rather than anchor decisions to dated failure-case studies, update evaluation suites quarterly, and prioritize task-specific fine-tuning where newer models show measurable gains in reasoning and instruction-following (source: Ethan Mollick on X).