I used to develop FSD. My take: A lot of what they are saying are super interesting and makes sense. It seems that they had some really intelligent talented experienced hard working engineers given freedom to do what they thought was the best solution. This is pretty rare as an engineer. I assume a lot of engineers around the world working on the same problem will think ”This is what I think we should have done” on various parts of the problem.
One detail I really liked that I haven’t seen commented on enough is what Karpathy said. First you do an initial attempt. Then you identify a problem. Then you add it to your unit tests. Imo this is the big differerence between a great engineering approach and what many other teams are doing. You create unit tests and make sure that every commit improves on unit tests and you make the unit tests tell what is a good solution.
One detail I really liked that I haven’t seen commented on enough is what Karpathy said. First you do an initial attempt. Then you identify a problem. Then you add it to your unit tests. Imo this is the big differerence between a great engineering approach and what many other teams are doing. You create unit tests and make sure that every commit improves on unit tests and you make the unit tests tell what is a good solution.