This talk will cover best practices in active learning with real-world examples. How search engines use active learning to select the most impactful results to show raters to improve relevance. How self driving car companies can guess the answers and show them to annotators to get a 10x speed up in data collection. This talk will also go over the state of the art of transfer learning. How LSTMs can be trained on one language and applied to another. How image-net was collected and drove the growth of vision algorithms. How neural nets make it easy to fine tune existing algorithms and get great performance with a small amount of training data. This talk will also cover best practices in training data collection. Training data collection strategies are often overlooked but are often the difference between a successful AI deployment and a science experiment.