[cityhall]

MNIST error rates are noise these days, but otherwise agree. These guys have been around for the entire deep learning boom of the last decade, and have never produced results to match their handwaving predictions. They're the Moller Skycar of the field.