Ofcourse its gonna improve. The real question is what is the rate of improvement versus other AV systems. But based on what we are seeing their rate of improvement is akin Nissan and not Google. They have a very average but subpar rate of improvement.
Tesla fans always claim this exponential magic rate that is 100% fabricated and doesn't exist. What tesla fans should be praying that their rate of improvement is atleast as good as Waymo.
Waymo
2009 - 25 mile (based on their 10 100 mile uninterrupted challenge)
2012 - 500 (guesstimate)
2015- 1,244.37 mile (CA Disengagement report)
2016 - 5,127.97 mile (CA Disengagement report)
2017 - 5,595.95 mile (CA Disengagement report)
2018 -11,154.3 mile (CA Disengagement report)
2019 - 13,219.4 mile (CA Disengagement report)
2020 - 29,944.69 mile (CA Disengagement report)
2021 - 35,000 (projected based on history)
2022 - 90,000 (projected based on history)
These are a useful starting point, but one has to normalize for the quality of those miles, vs both where Tesla vehicles are operating and even changes between years for Waymo.
For many of those years, most miles accrued were probably in Mountain View, CA. I worked in there much of last decade, certainly some challenging intersections but nothing crazy. But at some point I assume they started doing more testing elsewhere, was that a steady increase or step function increase in a few years? For instance, it looks like not much improvement btw 2018 and 2019, but if Waymo tested in more difficult areas in 2019, then that would be considered bigger progress than it appears.
With Tesla we have no real data, we have to do some estimation for any comparison of absolute state and rate of change. I really don't know about absolute state. I like to think of what it would be in one particular area, say Mountain View, or Chandler Az.
If FSD 10 was operating in Chandler, what do you think the miles / disengagement would be? I'd guess somewhere in the 100s, maybe. Maybe less.
So quite low compared to Waymo. That being said, the rate of progress in the past year seems faster than anything Waymo has showed. I could easily see a 10x on this metric in the past year.
Based on so much of the interventions being routing confusion, I could easily see this improve 10x in the next year. At that point, Tesla could be in the thousands of miles / intervention (again I'm saying on the same distribution of environments Waymo is reporting on, not just downtown SF).
So Waymo is probably 100 times better than Tesla right now, and maybe 10 times better next year. After that I don't have a clear idea of where the interventions are going to cluster and what the limiting factors will be. How much more can Tesla improve at that time, and how quickly. I would guess the rate of improvement would slow down a lot. Maybe 2x or 3x. Much is dependent on how many of the issues are due to limitations of the cameras and inference compute power.
But if Tesla can reach 10x from current state, I think that makes a darn good L2 system. Enough to make most customers happy. Progress to L4, I'm gonna reevaluate in another year.