Well maybe they should find better training methods or architectures