We arrived at our vacation home yesterday. Downloaded FSD 12.3.3.
Just used it to go to Costco, Trader Joes, and then home. Not an easy drive. Difficult left turns, roundabouts, speed bumps. It did it 100% robotaxi level with no interventions. At my house it pulled over and stopped (didn't go into the driveway). My mind is blown that we are down 40% this year on less auto growth when FSD is very close to being solved.
I sold 172.5CC for next week at $1 earlier today. I couldn't get myself to pull the trigger on 160 strikes or lower. With the stock down so much this year going into ER, I wonder if we actually go Green next week. Who knows.
I am equally blown away by just how good the current FSD is. I found the Auto Max Speed setting particularly good - great job of choosing the car's speed most of the time.
I agree that in the march of 9s, this edition of FSD is a big step forward towards robotaxi. I do not agree that FSD, or at least FSD as something more than supervised driving assist, is very close to being solved. Though this might be a terminology thing - there is solved as in it does most everything necessary but in a "lab" or restricted setting. When I hear 'solved' that sounds like we're ready for the regulators, business model, and other issues having to do with adoption and deployment, rather than significant remaining technical development reliability improvements.
I've also worked in a closely related field, and I've seen models that reached the right stage of development that just take off in how fast they improve in quality and usability of the output. These can easily be a 99% effort for 20% of value (the data management task of building the infrastructure and data structures and on and on that collects and labels the data, the processing engine that can train and retrain on all that data, etc..) followed by the 1% effort (training on that data, and evolving the models) for 80% of value. In these situations there's no meaningful discount on the data management work - you gotta do it all to get to start on the 1%. Shortcuts are how people get a public demo from a technology that has no hope of development into a solution.
All that being said, I don't see FSD (or at least robotaxi) as close to being solved. If we decide that robotaxi doesn't need to navigate residential driveways and stuff - just needs to be able to pull up and stop in places that Lyft does, then it actually seems like the current version is headed in that direction. I've even had it navigate a Whole Foods parking lot and pull up to the front door as if it were dropping me off there.
That's more evidence of getting closer.
If we routinely notice the improvements from version to version, then we're not there - not for robotax. At least months after we've stopped noticing the release to release improvements, with our perception being of really high quality - a period of time in which the invisible march of 9s is going on - that's when robotaxi is getting closer to the actual monetization efforts (building vehicles, a network, regulators, ...).
Of course - we're not trading the company and the technology. We're trading the broad market view of the technology and its monetization. How do investors view the technology? An important reason why I see a ChatGPT moment being hard to come by - there are maybe 100k people in the US that can have and do have personal experience with FSD. That's just not enough people for a ChatGPT moment.