EVNow
Well-Known Member
What I'd like to find out is - how are they prioritizing to deal with scenarios causing disengagements. Is it all manual ? Can they automatically find out what scenarios cause most disengagements ?Their approach seems endlessly flexible. That doesn't mean "ready soon", mind you. But it means that it's a practical approach.
For eg., if they get 1M disengagements a day, obviously they can't manually deal with them. Infact they can't probably deal with anything more than a hundred a day.
Once they find the top problematic scenarios, do they have to manually select the training set ? This would be not as bad as manually finding the important scenarios - but this is going to dramatically slow down the bug fix rate. Then they have to figure out whether heuristics need an update, which would be manual.
In other words, how much of this whole march of the 9s is automated vs manual. The more manual it is, the less exponential the pace of solution.
Ofcourse, currently they are just gathering scenarios / training data to be feature complete. I'm really curious what those set of features are … but that's a different story.