Many people have been trumpeting "superhuman" performance of deep learning on narrow tasks, and extrapolating that to a future where self-driving cars are ubiqu

Many people have been trumpeting "superhuman" performance of deep learning on narrow tasks, and extrapolating that to a future where self-driving cars are ubiquitous and saving our lives routinely. That performance is technically true, but unfortunately it doesn't translate to real-world engineering problems like driving cars.