I was a HUGE shill for Tesla's vision (no pun intended) of full self-driving. However, between the (IMO) extremely short-sighted decision of removing ultrasonics to cut costs ahead of Vision replacing them, Elon's ridiculous antics over at Twitter, unjustifiable price increases of both their hard product and FSD, and, most importantly, FSD Beta being downright dangerous to operate, my faith in Tesla solving autonomy has all but evaporated.

It wouldn't surprise me at all to hear that Tesla is doing funny accounting with their Autopilot crash data. I'm starting to think that the feds will drop the hammer hard on their autonomy pursuits in light of this and will force them to:

- Change the product name to something less misleading (While I'm okay with ambitious product names like this, SO SO SO MANY people are not), and

- Honor customers wanting a refund (and I bet there is a long line of people that want one)

I've sold all of my shares in Tesla (admittedly, I didn't have many) and have not only cancelled my interest in another Tesla but have considered trading in our Model 3 for another EV despite a smaller charging network multiple times this year. (To me, there's no point in owning a Tesla if their supercharging network will be forced open to everyone and their path to autonomy is infeasible.)

I've been using FSD Beta in our Model 3 this past year. It has gotten a lot better throughout that time frame, but it is very far from safe to use, and I certainly wouldn't recommend it for anyone but the faint of heart.

When it works, it works really nicely. Beta has smartly routed around construction and does a good enough job of navigating around other cars and predicting their behaviors.

Unfortunately, when it doesn't work, it REALLY doesn't work.

Even in the most recent release that dropped this month, Beta has attempted to run red lights and drive on wrong sides of the road, has missed or made wrong turns a surprising number of times, and done some of what I can only call "generally extremely wild shit" (like super sudden stopping and sharp turns without telling you why). It usually picks lanes correctly, but will just as often pick lanes that will take you to the wrong place or straight up don't make any sense. It also LOVES switching lanes at the very very last minute, usually right before a turn onto another road, which is confusing for everyone.

It's actually gotten worse in situations previous versions excelled in.

I even turned it off completely at one point because it was about to miss a turn that it nailed several times before, in broad daylight, with no other traffic around, for no reason.

Recently, I've been thinking about GM, Ford and Waymo's approach to autonomy (LiDAR for object measurement and detection and pre-mapping as many roads as possible). If Google Maps could map over 80% of America's roads within two or three years, then assuming that mapping can be done "at the edge" (i.e. with regional mapping vehicles), what's stopping these companies from doing the same thing? Furthermore, if mapping and equipping cars with LiDAR scales well, then what worth is there in trying for a vision-only approach?