jeewee3000
Active Member
I'm on the other side of the fence: L5 FSD is inevitable and will come earlier than the masses expect it to arrive. (Most say: at least five years from now, or more, or never).This video is excellent in showing the need for much more improvement. I still have doubt anybody can do the FSD level 5 defined as it can do ALL cases like a human can, like an accident on freeway or street or police pursuit with road closure. Have to see to believe.
First off, your doubts are underestimating what deep learning can do given enough high quality data. The improvement will grow exponentially once DOJO arrives (unsupervised labeling -> much faster learning cycles).
Second, the problems you discribe: an accident, a police pursuit, ... are all just ballistic problems. The car will see the same a human sees and in most cases much quicker than a human can. If a human is surprised by something and crashes (for example a tree falling on the road just as he is driving there), the FSD-car might also crash, but these are edge cases first and foremost AND even the human couldn't prevent the crash. Therefore: FSD was not more unsafe than human driving. (Also, the FSD-car will brake sooner and harder than the panicky human, trying not to skid (and to prevent sideways impact) resulting in lower impact damage and lower injuries.)
Third, and this point feels like beating a dead horse, FSD does not and will not result in 0 accidents on the road in the future. The simultaneous surround-video at all times, without a millisecond of inattentiveness, combined with not breaking the speed limit, will already erase the vast majority of accidents out there. Even if in some rare edge cases the FSD-car reacts in a more stupid manner than the human and causes a crash, statistically this will be just a blip on the radar overall.
The metric therefore should be along the lines of: [total amount of damages in crashes] per mile driven. This is however impossible to calculate so we'll just have to work with [amount of crashes with fatalities/injuries/no injuries]/mile driven. FSD will crush this too. (Just erase the death toll of drunk driving alone....)
Current FSD beta might already beat humans on that metric. But this fact will become more and more blatantly obvious in the coming months/years. Therefore I can't wait till FSD beta becomes a public release and Tesla Insurance can report accident data that will shock Insurance companies. From that point on, it's only a matter of time before the regulators follow.
So yes, I'm very bullish on FSD. I'm an attorney dealing mainly in bodily injury (be it by car accidents, work related injuries or other) and 99% of automotive injuries are due to someone not being aware enough of their surroundings (whatever the reason). In 50 years time our descendants won't be able to fathom our time allowed 'human driving'. "What? That sounds so dangerous, grandpa!" - "You think that's bad, [insert errors of humanity from the 20th century]"
The world will become safer, more fun and more productive with FSD. And us TSLA shareholders will make a nice profit on the dawn of FSD as well. (A unique moment in history, I might add, truly groudbreaking. I can't hide my excitement about all this.)
Cheers to Tesla, DOJO, FSD and the disruption they will bring.