[Innovation is being completely mispriced in the public markets as the public markets go passive]: I totally agree. John Bogle [the inventor of the index fund] was awesome. May he Rest In Peace. One of his last comments was that passive index funds are too big. [You can learn more about John Bogle by listening to Founders #57 and #58]
[We believe that due to cost declines in battery pack systems, we will see EV sales increase from 1.3 million in 2018, to 26 million by 2023. A 20 fold increase if the battery capacity and production capacity is there. What do you think of that?]: Elon’s response: That sounds about right. You might be off by a year or two but not by much.
I want to emphasize that if I give estimates there is a lot of guess work. Especially on an exponential curve – a year or two difference is enormous. We got a lot of criticism for the number of cars we delivered in 2017. The area under the curve of production in 2017 was quite small because it was the beginning of an exponential ramp. Once that got going, the area under the curve was enormous. People were so shocked. Last year we made about as many cars as we did in our entire history [combined].
[Most people thought it was impossible]: If you were doing a linear extrapolation it certainly would be.
When making estimates against an exponential curve, small changes in the calendar breakpoint have enormous percentage differences. The time difference is small, but the percentage difference is enormous.
My estimate of getting to 5,000 cars per week was off by approximately 6 months. In the grand scheme of things, 6 months late for a massive new program is not too much. This was characterized in the press in terms of the percentage of units instead of a calendar shift. It was perceived as a massive shortfall when in fact it was merely a 6 month delay.
[With that in mind] If you are asking what my guess for production for Tesla in 2021 is, it’s 1.5 million cars.
You can think of our cars, long term, as carriers for autonomy software. They are a vehicle, literally and figuratively, for autonomy software.
I still think the last 10% of autonomy is extremely difficult. Or even the last 1% of autonomy is really difficult.
I think we will be feature complete, full self driving this year. Meaning the car will be able to find you in a parking lot, pick you up, take you all the way to your destination without an intervention. I am certain of that. That is not a question mark.
Some people will extrapolate that to mean it works with 100% certainty, requiring no observation, perfectly. That is not the case.
Once it is feature complete then we have to figure out how many 9’s of reliability do you want it to be? And then when do regulators agree it is reliable? I manage autopilot engineering directly every week, in detail. I’m certain of this.
Something we don’t control: When will regulators allow these features to be turned on with human oversight? And then when can this be done without human oversight?
Some people think that I am a business person, or a finance person. I’m an engineer. I do engineering. Always have. I wrote software for 15 or 20 years. I understand technology and software at a fundamental level. I think we have an extremely good technical team. I am certain we will get this done this year.
The advantage we have that is very difficult to overcome is we have a vast amount of data on interventions. Effectively, the customers are training the system to drive. There are millions of corner cases that are so obscure and weird you wouldn’t believe it.
Every time somebody intervenes, takes over from autopilot, that information is saved and uploaded to our system.
The reliance on LIDAR is unwise. It gets you to a certain point but no further. A series of if/then statements and LIDAR will not solve it. You have to solve vision, perception, essentially understanding. And then it is solved. You don’t need anything else.
[How Elon describes how humans drive cars now]: We drive cars with basically two cameras that aren’t very good. On a gimbal that doesn’t move very fast.
If you want a complex neural network you need a combination of software and hardware. Your software needs to be better to compensate if the hardware is weaker. Think about how video games progress. It’s a combination of software and hardware. No amount of clever software could produce a video game of today on old hardware. It is the same for neural nets.
We started off with highway [autonomy] because that tends to be what matters most. Stop and go traffic is painful. Freeways are usually congested in every city in the world. Highway accidents tend to be higher velocity and more dangerous. Fatalities are very much related to speed.
When I hear companies say they are changing the world I think of Silicon Valley [on HBO]. The show is pretty funny. Literally every company says that. Making the world a better place by social/mobile/crypto.
