I have several questions about Karpathy's presentation: 1. How to label automatically using fleet learning? An example he gave is detecting cut-ins. However, I

I have several questions about Karpathy's presentation: 1. How to label automatically using fleet learning? An example he gave is detecting cut-ins. However, I am wondering if it could be applied to more general and basic cases, for example, labeling lane marking/vehicles. 2. How do you know if someone is a good driver (which is essential for imitation learning)?