A new frontier in AI research focuses on incorporating more human evaluation to critically assess the performance of AI models, ensuring they meet real-world effectiveness despite surpassing standard benchmarks.