Here is how I view $TSLA, Tesla, the company has demonstrated capabilities in EV, and in battery storage, ability to build big factories, deliver cars. Now, they have promised a big future in robotaxi, and robotics. Musk has shown consistently, ability to work technology miractle, and until very recently able to get a cult like following, which helped him to finance, whether at Tesla or at his other ventures.
However, existing $TSLA business is worth hardly 5% to 10% of its current valuation and the 90% of the current valuation is based on future potential. This is what gives me pause. The probability of Tesla realizing this valuation and growing further is very low, it doesn’t mean they will, or will not be able to achieve their goals.
It is about whether I need to bet on such a low probability outcome at these valuations? I don’t mind betting on some low probability high payout scenario. It is not clear to me we are looking at 5x, 10x here.
We can debate whether tesla has superior technology, whether Tesla can replace Uber, whether they can solve regulatory challenge across the states overnight, etc. But at this point it doesn’t make sense to invest in those debates because current valuation doesn’t make it worthwhile.
Of course. I just watched 35 minutes of netflix in the car while my wife was shopping! Also, on my phone, if I’m suspicious about anyone hanging around near my car, I can connect to the car and watch simultaneous live video from multiple cameras on the car to see what’s going on.
Fair comment in a nod to realism, but the point and reasoning of the illustration remain, and the best outcome could still be (1) or (3) (if we ignore LIDAR only, as you say) and still Team 2 has the more diverse outcome set (two vs one possible outcome for success or double the opportunity set for success relative to Team 1).
By diversified, I am referring to the number of possible outcomes that can have success. Team 2 has the higher number.
If one desires more realism in the example, one could modify Team 2 to have Camera, LIDAR, and RADAR (as one example).
And then modify the outcome set to include all combinations of all 3 sensors types (now 7 possible outcomes).
Other possible configurations could include number of sensors, their placement, etc (this might complicate the outcome set, but this is the scenario analysis that the AV companies face when making strategic technology decisions).
The point is to consider which technology covers which range of possible future outcomes.
It’s interesting to note that the Chinese AV companies are similar to Waymo - adopting more different types of sensors relative to a camera-only approach like Tesla.
So if we look at what AV companies are actually doing, most are opting for a “hedge your bets” approach, a Team 2 approach, versus a “put all eggs in one sensor basket” approach, a Team 1 approach.
Apparently the next FSD release will have 4x the parameters (weights) of the prior release.
So this AI is increasing in complexity. The camera-only brain is getting 4x bigger (when measured by number of parameters).
What’s interesting is, for a given level of functionality, you want fewer parameters (reduced complexity, reduced computations, higher precision in parameter estimation).
But the tech is not yet at this optimization stage, it is still at the stage of “can we find any sensor array and AI, within our budget, that can deliver AV functionality.”
Tesla’s have been sending videos from cameras to their servers for about 10 years AFAIK.
And just for reference, Facetime (using cellular frequencies) varies from about 1 to 3 Mbps. The 4G LTE cell towers support 50 Mbps (phone/car) data upload.
I mention Facetime, because it needs to be low latency as well as handling the bitrate. Other uses like sending Sentry mode videos from cars to your phone can actually tolerate a few seconds or more latency delay. (even though the 4G spec is something like 30-70 msec)
And 5G networks provide 100-300 Mbps upload with 1-10 msec of latency
“The stable genius #2, who confuses money with expertise, decided not to use LIDAR.
We don’t know the preferences for LIDAR of the actual technical experts at Tesla.”
We will find out, eventually, if this becomes another sorry example of management incorrectly making technical decisions that are outside of their knowledge domain.
It’s noteworthy that the other competitors use camera plus LIDAR. So Tesla decision makers are exceptional in some regard, either more in the success direction or more in the fail direction.
Side note: My understanding is Tesla uses LIDAR for some smaller scale testing. This means Tesla has a large camera-only training data set and a much smaller camera plus LIDAR training data set, so Tesla does place some degree of value on the sensory data provided by LIDAR.
IIRC a year or two ago this was discussed and Tesla said that they are always testing against camera+LIDAR to see how the vision models compare to camera only and to see how LIDAR tech and costs compare as well. They were a big customer of Luminar.
Skimming through this thread, I do find it slightly comical - the amount of time (years and years) and money (millions and millions? Billions?) that have been poured by the worlds biggest companies into getting the fanciest computers and brightest engineers - to do what we can teach basically any 16 year old to do with a few intermittent practice sessions over a month.
I thought technology was suppose to help us do cooler and better things, not a more complicated way to sweep the floor or mow the grass
Yes, but the 16 yo wants to be paid and to have benefits like sick days, paid vacation, heath insurance, retirements. Often those benefits are 30% of pay.
Once fully depreciated, robot works full time w/o complaint. Only requires energy and maintenance. A much better deal.
