This is the part I can't understand. You release software to dogfood, public beta, etc when you want to get more feedback and you've exhausted the ability of the previous tier to provide it. You wouldn't release software to dogfood if the programmers on the project can readily find defects in a minute or two, you go to dogfood if the engineers use it all the time and can't find new defects. The only reason for this FSD software to be in the wild is it is part of the larger pump and dump scam.

It would take too long, on the contrary you want to release it as early as possible. It need to learn from real life situation. To get a true real life data you have to... release it to real life situation.

>Why is the car stopping in the middle of nowhere?

garbage collection?

I think is because they're going for a full machine learning approach, the neural network tries to learn the best solution but is unable to receive simple instructions like "go in a straight line"

They're not going for a full ML approach. That's just more Elon marketing language. Tesla reverse engineers have gotten their hands on the FSD beta and inspected the internals - they use ML for perception and scene understanding, but all of that information feeds into traditional robotics prediction, planning, and pathfinding algorithms.

> You have thousands of engineers to throw at the problem - why can't you stay in a lane while turning? Why can't you read turn arrows in the road? Both of these are extremely fixable problems. Why is the car stopping in the middle of nowhere?

Why would anyone risk blowing that whistle when Musk and Tesla have tried to get employees murdered by the police by accusing them of being mass shooters[1]?

[1] https://www.bloomberg.com/news/features/2019-03-13/when-elon...