Here’s an interesting video presentation by Andrej Karpathy, where he extends the idea of Software 2.0 and shares several practical lessons learned at Tesla.
This video actually left me more inspired to learn how to do machine learning. Some topics:
The evolution of software from domain-specific engineering, to an automatic search through the space of programs.
The much higher importance of creating, understanding, and refining training data in real-world applications vs. academia.
That the very rare training example is sometimes the most important to get exactly right, e.g., car vs. trolley, mispainted lane lines, blue traffic signals.