“Brandon Foodi predicts that a significant portion of knowledge work will shift towards humans creating evaluations ('evals') for tasks that AI models cannot yet perform, as this is more efficient than humans repeatedly performing the tasks themselves.”