<>
.uix_logoSmall img {
max-width: 290px !important;
max-height: 100px !important;
}
@media only screen and (max-width: 370px) {
#top_right_logo_img {
display: none;
}
}
-
Want to remove ads? Register an account and login to see fewer adsand become a Supporting Member to remove almost all ads.
TeslaTSLA & the Investment World: 2025
- Thread starter doug
- Start date
- Status
cliff harris
Active Member
I'm no expertbut I am an ex AI coder from ages back and think about this a lot
The thing iswhen you go end-to-endyou are basically ignoring a lot of data that you might *think* would be helpful. End-to-end NNs means that the car is literally driving like a human. I drive a model Ybut have no idea how wide it isor how long it isor what its exact turning radius is. The way I drive it is the way any human drives a new car when they buy it: They drive VERY carefullyand gradually push the limits until they get a near miss with a bumper or a wing mirrorand at that point you have 'learned' how wide and long your car is. You probably also make use of gauging shadows and reflections etc.
I doubt that there is anywhere in the FSD code where it is helpful to just plug in the dimensions of the vehicle. Ditto the stopping distances at speedsor even the turning circle.
In theory you could stitch together a whole bunch of neural networks and have some glue-code that links them and gets involved in the decision makingand MAYBE you could include a system for specifying vehicle data there. To some extent Tesla MUST be doing this to ensure the vehicles obey laws to the letter where humans do not. I suspect that the actual arrangement of code is a bit hacky and a bit of a messas all large codebases become...
So TL;DR: The cybertruck was probably running standard FSD in shadow mode for a whileand doing so cautiouslytesting the limits. I suspect Tesla want to first check no weird anomalies are presentand also only enable FSD for a vehicle when they have 10,000+ miles of shadow FSD metrics to sanity-check against.
NoTesla never lost near as much money for equal volume of sales. It was never anywhere near as bad as Rivian's losses
Elon said "Robo-Taxi hardware capable vehicles" or something to that effect. That also ended up not being true but still different than what you imply
In the long-term yesbut realistically the order that keeps the company alive to get to this FSD dream is: carsmegapackspowerwallsand then FSD software sales.
I'll go on record with others that againFSD won't be ready in 2025. If it isI'll be selling my shares and gone from the site!
Happy New Year!
FSD won't be ready in 2025? What do you mean by "ready"?
BobbyKings
S85D > M3P > S Plaid
Is the title of this thread referring to a best before datean expiration date or an it will get better soon date?
I'm speculating a new thread will be opened and it will be like Groundhog Day movie and all the same talking points get repeated again.Is the title of this thread referring to a best before datean expiration date or an it will get better soon date?
I didn't disagree with your postas I think the intent of what you said is valid: Tesla has demonstrated many times they are willing to make the decision to change course despite needing to start from at/near square oneabandon plans/effort/sunk costtake on a new roleetc...
But I think to suggest it hasn't been difficult doesn't acknowledge that doing so (often successfully) is what sets Tesla apart. Making the decision no doubt has been difficult... and then executing on that decision is a tremendous effort... the fact that from the outside it often doesn't seem so is a testament to how hard they work.
Seats not up to par? We'll make out own.
One of the leading autonomous system OEM's equipment/offerings not up to snuff? We'll not only build our own systembut we'll spin our own custom neural net CPU ASIC's to do it.
Availability of compute for model training not available/cost-effective? We'll build our own supercomputer
Electric power steeringsteel panel forming equipmentcell manufacturesingle piece casting processesetc... the examples go on.
Soin essence I agree with youjust wanted to point out that indeed it is precisely the difficulty in pivoting that makes it so amazing.
... but.... can't the stealing Chinese steal the curated and trained data directly from Tesla? They have done it with impunity in the past. Cisco and Tesla comes to my mind.
Why steal the raw material when you can steal the finished product.
Instead of all the pie in the sky about FSDwhy aren’t we estimating the quarterly/ yearly sales numbers and trying to figure out sales/ income for the year? You knowsomething real?
Thumper
Active Member
Thank you GGR for all your work over the years. Wish you the best!
Todd Burch
15-Year Member
In theory yes. But I don't think stealing petabytes of data is an easy task. Tesla is not foolish enough to make it available to a public networkand you can't walk into the server room with a 50 PB jumpdrive
,
Todd Burch
15-Year Member
Not saying that pivoting is easy. Just that Tesla has no problem committing to a decision to pivotvery much because Elon is acutely aware of sunk cost fallacyand his DNA is built on the concept of "fail fastfail oftenand iterate."
Excellent work by Jordan as usual. If you want to know why the Cybercab only has wireless chargingthis has many answers. It demonstrates how thoroughly Tesla has thought about the overall cost structure of the robotaxi business.
Last edited:
Nah. The real experts we have floating around on the forum will tell you Tesla has no real cost or scaling advantages over the likes of Waymo.
DFibRL8R
Active Member
Apparently someone blew up a Cybertruck in front of Trump Hotel in Las Vegas:
- Status
Similar threads
- Marketplace listing
- Locked
- Marketplace listing
