he more you train deep learning models, the more you realize that there is still so much left to figure out with NNs..

Neural networks still

- require a lot of data cleaning
- need a fair bit of featurization
- regularly overfit
- often don't learn the nuance in the data