Unawareness of Deep Learning Mistakes

TL;DR: People are hardly aware of any deep learning mistakes they made, because things always appear to work, and there are no expectations on how well they should work. The solution is to try to accurately reproduce settings & performance of high-quality papers & code.