5 ms·

I think the argument for it goes as folllows: Tesla has more 'self supervised' driving data than any other car brand, so if data is key to making self driving

I think the argument for it goes as folllows: Tesla has more 'self supervised' driving data than any other car brand, so if data is key to making self driving work, they win. You could also see how this is a winner take all monopoly once self driving gets cracked. Now before I get flamed, I don't own Tesla stock, and no I don't think self driving is just around the corner. I'm perfectly happy with good autopilot though and I think the model 3 is a great mid level car. But yes I think Tesla stock was quite over valued.

rsynnott 3y ago
> You could also see how this is a winner take all monopoly once self driving gets cracked. … _Why_, though? Even if you buy the idea that Tesla is going to come out with self-driving cars Any Day Now(TM), that really seems like the underpants gnome part of the narrative. It’s most unclear why it would be a winner takes all thing.
- Ekaros 3y ago
  It is only winner takes all if Tesla keeps being only one with self-driving... Which to me looking into history is insane. Is there any technology where there is only single provider? It is big enough market that eventually there will be multiple-players and then it will be race to bottom. Marginal profits. Self-driving taxis will be dumped like Uber and Lyft did against other players. And then it will be low margin business.
  - gravescale 3y ago
    Right? And even if they did score a virtual monopoly in the US through assiduous lobbying, they can't keep a global lid on development forever. Especially that if they do crack it, they prove that it both can be done and it's "just" the software. Even without the Asian manufacturers, the German ones wouldn't just roll over and die either.
- VirusNewbie 3y ago
  It’s winner take all because if the scaling hypothesis applies to self driving, they have way more telemetry to do self supervised training on, which initiates a feedback loop: more people buy their cars which gives them even more training data, making their self driving system even further ahead of the competition.
  - gravescale 3y ago
    Even if that were the case (big "if"), that can't go on forever. Even if it's literally just a matter of who has enough data and training core-years, at some point a competitor, maybe with state-level support, will have "enough" data to come after you with their own self driving option. Perhaps now Tesla is at 90% (just say, it's just an example) and Mercedes (again say) is at 10%. One day Tesla may be 100% and Mercedes is at 30%, and the German government, for one, will be shoveling cash into the furnace the whole time if the German auto industry is in existential danger. The gap can only close. Meanwhile once you reach full self driving, you're stuck at diminishing returns. Doubling your data won't double that: at best you could make a very marginal improvement to safety stats or range economy. Maybe with a lot of lobbying you could swing a slightly higher Tesla-only speed limit, but double self driving isn't a thing. Everything else in the car is commodity stuff like sound systems and seat heaters that anyone with a car factory and a team of designers and a copy of the Bosch Automotive Handbook can do if they want to. Probably you could get some network effects from the "Rent your Tesla out" app, but again, that's an option for anyone with a team of designers once the self-driving part is done and they'll be in the disruption phase and aggressively undercutting as best they can. This also assumes that in 2029 you can't just use 80% (again... just a random date and a random number) synthetic training data and leapfrog nearly all of those hard-won physical miles. The costs of doing that will plummet too, so the training cost moat will only shrink as well. Meanwhile Tesla will be ossifying as gigantic trillion dollar behemoths tend to do, accruing costs like pensions, management entrenchment will be in its third decade and so on. On the other hand, if we're doing hypotheticals, if it did come to pass and Tesla is sitting pretty on a 13-figure valuation predicated on it being the only self driving car out there and a competitor pops up suddenly from stealth mode with a fully-formed alternative, much of the value could evaporate so fast it would be a structural hazard to the economy. That would be an interesting situation. And, remember, are that point it's all predicated on the software being the key, so Upstart Autonomy Inc. can license it to Mercedes and get it in cars almost immediately. Is Tesla worth a lot? Undoubtedly. Is it that much? Questionable to me. But then the market disagrees, so what do I know (but also the market was off by 30% by it's own estimation so how much does it know?)

rsynnott 3y ago

> You could also see how this is a winner take all monopoly once self driving gets cracked. … _Why_, though? Even if you buy the idea that Tesla is going to come out with self-driving cars Any Day Now(TM), that really seems like the underpants gnome part of the narrative. It’s most unclear why it would be a winner takes all thing.

Ekaros 3y ago
It is only winner takes all if Tesla keeps being only one with self-driving... Which to me looking into history is insane. Is there any technology where there is only single provider? It is big enough market that eventually there will be multiple-players and then it will be race to bottom. Marginal profits. Self-driving taxis will be dumped like Uber and Lyft did against other players. And then it will be low margin business.
- gravescale 3y ago
  Right? And even if they did score a virtual monopoly in the US through assiduous lobbying, they can't keep a global lid on development forever. Especially that if they do crack it, they prove that it both can be done and it's "just" the software. Even without the Asian manufacturers, the German ones wouldn't just roll over and die either.
VirusNewbie 3y ago
It’s winner take all because if the scaling hypothesis applies to self driving, they have way more telemetry to do self supervised training on, which initiates a feedback loop: more people buy their cars which gives them even more training data, making their self driving system even further ahead of the competition.
- gravescale 3y ago
  Even if that were the case (big "if"), that can't go on forever. Even if it's literally just a matter of who has enough data and training core-years, at some point a competitor, maybe with state-level support, will have "enough" data to come after you with their own self driving option. Perhaps now Tesla is at 90% (just say, it's just an example) and Mercedes (again say) is at 10%. One day Tesla may be 100% and Mercedes is at 30%, and the German government, for one, will be shoveling cash into the furnace the whole time if the German auto industry is in existential danger. The gap can only close. Meanwhile once you reach full self driving, you're stuck at diminishing returns. Doubling your data won't double that: at best you could make a very marginal improvement to safety stats or range economy. Maybe with a lot of lobbying you could swing a slightly higher Tesla-only speed limit, but double self driving isn't a thing. Everything else in the car is commodity stuff like sound systems and seat heaters that anyone with a car factory and a team of designers and a copy of the Bosch Automotive Handbook can do if they want to. Probably you could get some network effects from the "Rent your Tesla out" app, but again, that's an option for anyone with a team of designers once the self-driving part is done and they'll be in the disruption phase and aggressively undercutting as best they can. This also assumes that in 2029 you can't just use 80% (again... just a random date and a random number) synthetic training data and leapfrog nearly all of those hard-won physical miles. The costs of doing that will plummet too, so the training cost moat will only shrink as well. Meanwhile Tesla will be ossifying as gigantic trillion dollar behemoths tend to do, accruing costs like pensions, management entrenchment will be in its third decade and so on. On the other hand, if we're doing hypotheticals, if it did come to pass and Tesla is sitting pretty on a 13-figure valuation predicated on it being the only self driving car out there and a competitor pops up suddenly from stealth mode with a fully-formed alternative, much of the value could evaporate so fast it would be a structural hazard to the economy. That would be an interesting situation. And, remember, are that point it's all predicated on the software being the key, so Upstart Autonomy Inc. can license it to Mercedes and get it in cars almost immediately. Is Tesla worth a lot? Undoubtedly. Is it that much? Questionable to me. But then the market disagrees, so what do I know (but also the market was off by 30% by it's own estimation so how much does it know?)