The upfitting step at roughly $50,000 per car is the detail I hadn't seen broken out. If the fleet really goes from 4,000 to 20,000 by end of 2027, the vehicle mix by then looks like a bigger driver of unit economics than the headline count. Is there any data on utilization per vehicle yet?
Bringing costs down and increasing production is the next logical step.
That said, Waymo vs Tesla is a no-contest, despite what Tesla fans say.
Waymo has spent 18 years or so doing very diligent engineering. It adapted its platform as AI became more powerful and new techniques arrived, such as Transformers.
But it kept the modules very carefully compartmentalized, so a failed prediction does not propagate easily to a hallucinated decision.
Tesla believes you just put it all in the neural net and it will work. As the scale of deployment goes up, the gap will be impossible to hide.
"Think, for example, of the 122-day sprint to build xAI’s Colossus data center"
In AI, Musk ran into the same problem as with FSD, and same will happen with its humanoid robots. When massively scaling predictable hardware fails, he is left without options.
In your earlier post on Tesla, which is now starting to scale (at least in Texas), you commented that "We don’t yet have good data on the safety record of these vehicles, and we shouldn’t draw any firm conclusions about Tesla’s safety record until we do."
I'm curious to know more about this. Does the Texas regulatory setup mean that at some point we will get more information, similar to the exhaustive stack of incident reports from Waymo that (if I recall correctly) you and Kai read through earlier?
The upfitting step at roughly $50,000 per car is the detail I hadn't seen broken out. If the fleet really goes from 4,000 to 20,000 by end of 2027, the vehicle mix by then looks like a bigger driver of unit economics than the headline count. Is there any data on utilization per vehicle yet?
Bringing costs down and increasing production is the next logical step.
That said, Waymo vs Tesla is a no-contest, despite what Tesla fans say.
Waymo has spent 18 years or so doing very diligent engineering. It adapted its platform as AI became more powerful and new techniques arrived, such as Transformers.
But it kept the modules very carefully compartmentalized, so a failed prediction does not propagate easily to a hallucinated decision.
Tesla believes you just put it all in the neural net and it will work. As the scale of deployment goes up, the gap will be impossible to hide.
"Think, for example, of the 122-day sprint to build xAI’s Colossus data center"
In AI, Musk ran into the same problem as with FSD, and same will happen with its humanoid robots. When massively scaling predictable hardware fails, he is left without options.
This is great news! Happy to have Toyota going after this with deployment in Japan.
In your earlier post on Tesla, which is now starting to scale (at least in Texas), you commented that "We don’t yet have good data on the safety record of these vehicles, and we shouldn’t draw any firm conclusions about Tesla’s safety record until we do."
I'm curious to know more about this. Does the Texas regulatory setup mean that at some point we will get more information, similar to the exhaustive stack of incident reports from Waymo that (if I recall correctly) you and Kai read through earlier?