I am not dogging on Tesla, but my concern is how can you realistically build an AI system that is going to drive just the way everyone like it to?
You can't-- but there's already 3 different "style" settings.... and as the system gets better I expect there'll eventually be more customizations made available.
It won't "drive like" every possible individual person- and good thing, since many individuals are terrible drivers--- but if they ever get past L2 you'll care a lot less because you'll be watching a movie, reading a book, playing a game, etc... (or sleeping if they ever get to L4).
As a human we can anticipate situations better. For example, while the FSD make have access to all these cameras and have 360 degree view, it doesn't know that in order to make a left at the light ahead we need to start getting over two lanes to the left much sooner than if I let the FSD system do it. I mean I can see that there is a bus ahead and it is going to stop to pick up some people so I anticipate and get from around the bus much sooner, will FSD do such things? That is where I struggle with this system.
Up until fairly recently in its development, the AI/NN/ML stuff was
only used for perception (ie what do I see around/head/behind me).
All planning and execution was traditional code. So pretty "dumb" and rigid.
Only much more recently are they starting to move other parts of the stack over and away from fixed/traditional coding. As Elon himself as noted, this is going to have weird results in the early stages- including making some things worse initially-- but will be much better in the long term as the training and models improve.
The point of the non-fsd regressions show they don’t really do this. At least for the non-fsd parts. Since that’s an easier problem I find it skeptical that they do much on the harder fsd problem either.
Again though, entirely different teams.
There isn't one "software department" that is run by one guy at the top of a vast hierarchy like most traditional companies doing software, and using one standard set of operational rules.
One of the reasons Tesla has been as successful as it is is by having relatively independent teams, with flat management structures, working on different things with different rules.
This ALSO has some downsides of course- (see the V11 UI as an example)- so it might mean ONE team produces something less good than if they'd done it another way. But generally when breaking new ground the advantages are massively bigger than the downsides.