Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[petit_bateau]

It's a fool's errand if Tesla solves this first because they very much want to license this which is what legacy Oems really want anyways. None if them actually cares to do it themselves, and only small players out of China want to steal Tesla's code and try just to make their product relevant in a world that doesn't care about Chinese made cars.

This is not just intended for driving mere automotive vehicles.

The ability to navigate/predict the physical world, and to take instructions (written, spoken, gestured) from the human 'semantic' world is pivotal for a robotic future.

Coupling FSD/etc with semantic/language models is not just of value to automotive. Mastering this is existential for any nation that intends to remain (or become) a modern industrialised economy.

 

[EVMeister]

More than 150 car models are now too big to fit in average car-parking spaces, according to analysis ..... While the size of the standard car-parking bay has remained static for decades, cars have been growing longer and wider in a phenomenon known as “autobesity”. ...... There is growing debate about car size and road safety, after two eight-year-old girls Selena Lau and Nuria Sajjad died when a Land Rover crashed through a school fence in south-west London in July.

More than 150 car models too big for regular UK parking spaces

Fears over safety as analysis by Which? shows ‘autobesity’ epidemic means cars getting wider and longer

www.theguardian.com

Original source for those who want more data. The Model X was mentioned as one of the widest models tested, but surprising to me (as I used to own one) is that the I-PACE is even wider than that. Model S also mentioned as particularly wide. Model Y and 3 weren't listed as being too wide for the average UK parking space, but as us Tesla owners over here know it can still be tight even with a Model 3.

 

[Unpilot]

This is not just intended for driving mere automotive vehicles.

The ability to navigate/predict the physical world, and to take instructions (written, spoken, gestured) from the human 'semantic' world is pivotal for a robotic future.

Coupling FSD/etc with semantic/language models is not just of value to automotive. Mastering this is existential for any nation that intends to remain (or become) a modern industrialised economy.

THIS is the way.

Tesla with FSD and Optimus (and any other unknown at this time project that can use a generalized AI) is hard to put a valuation on.

 

[cliff harris]

Original source for those who want more data. The Model X was mentioned as one of the widest models tested, but surprising to me (as I used to own one) is that the I-PACE is even wider than that. Model S also mentioned as particularly wide. Model Y and 3 weren't listed as being too wide for the average UK parking space, but as us Tesla owners over here know it can still be tight even with a Model 3.

The X is insane for the UK. My S was way too wide, and got be damage bills to prove it. The biggest fail in the design of the Y is that its not narrower. I would actually have paid MORE to make the car an inch or two narrower. At least its shorter than the S!
Even if the cybertruck is made for our market (doubtful as tesla have openly declared they don't care about RHD now...), it would be absolutely unusable here.

 

[Cosmacelf]

For those wondering how big a lead end to end AI is versus other competitors. I have a Rivian and, first, all competitors self driving is highway only. No one is working on urban driving like Tesla has with FSD Beta. Second, with the Rivian, I caught it making a mistake that Tesla had circa 2016. Driving in the left freeway lane, the system can get confused about where the fog line is based on k-rails, repeated posts on top of the k-rails, etc. Using that as a marker, Rivian, at least, is 8 years behind Tesla.

Also Elon said they will be spending $2B this year and another $2B next year on AI hardware compute alone. Not many companies can afford that.

 

[GhostSkater]

IMO Tesla will at least look at what Lucid has done and see if any of it can be applied to CT.

I'm talking about motors and the drivetrain voltage here, not frontal area and vehicle shape.

I know, we've seen them testing an Air at Freemont

It's just for the comparison I did it doesn't make sense, if we want to figure out how much more efficient a Cybertruck could be than a F150, we compare the Mach-e to the Model Y, not a Lucid with Model S

 

[Stretch2727]

Interesting live stream. Elon is giving lots of info about v12. Zero c++ code, all neural net, trained by just watching video. No labeling. No explicit designation of what a traffic light is, what a lane is, what a roundabout is, etc. it just learns what to do.

Reading all the posts on how v12 is all neural net from training video got me thinking it may finally drive like I do. Besides the obvious errors I don't really like the way FSD drives. Good drivers pay attention to surroundings and adjust. I will slightly slow down anticipating a situation where I may have to stop, position the car it a certain place depending on traffic etc. Although FSD is not doing anything wrong I find it hard not to interrupt if is going to fast or too slow or not in the right place on the road.

So everyone needs to drive like me so it is trained properly. Thank you.

 

[dhanson865]

I wonder what they'll do with Buffalo. They were meeting the employment requirement with thousands of labelers.

It took me some work to find it but I finally did

"Under this agreement, we are obligated to, among other things, meet employment targets as well as specified minimum numbers of personnel in the State of New York and in Buffalo, New York and spend or incur $5.00 billion in combined capital, operational expenses, costs of goods sold and other costs in the State of New York during the 10-year period beginning April 30, 2018. On an annual basis during the initial lease term, as measured on each anniversary of such date, if we fail to meet these specified investment and job creation requirements, then we would be obligated to pay a $41 million “program payment” to the SUNY Foundation for each year that we fail to meet these requirements. Furthermore, if the arrangement is terminated due to a material breach by us, then additional amounts may become payable by us."

from the TSLA 10Q

So they need to keep employment up there until 2028 at least. My thought would be to ramp up a battery cell and/or powerwall/mega pack site.

Not all the lablers would be able to tansition to the new jobs but you could at least hire new people and keep the jobs in total above the minimum.

 

[NicoV]

Also Elon said they will be spending $2B this year and another $2B next year on AI hardware compute alone. Not many companies can afford that.

Also, remember that Dojo started to process production workloads a couple of weeks ago. Tesla will (should?) get more compute/dollar from Dojo than the competition (using NVidia), so the bar for competition will be correspondingly higher than that $2B.
Not that that matters, because the competition doesn’t have the training data anyway.

 

[unk45]

Given that cars are not normally transferred between countries...

That premise is false. Throughout Europe cars routinely cross national boundaries. In the US most top medical personnel in Detroit cross the border with Canada twice a day, In Texas cross between border with Mexico every day is quite routine. Factually the fundamental issue is how the NN's can emulate a human who is familiar with those idiosyncrasies but with only visual cues.

Even the concept is difficult to understand for most of us. Decades ago I was involved in state-of-the-art AI, back when they were "expert systems". The advances in Neural Networks have become so rapid and excellent that human-level reasoning is limited by comparison.

Every day I use Google Translate or another; they are now better than any human aides we sued only two years go. Now we look at traffic behavior as yet another topic that seems, for the first time, actually on the cusp of success.

My remaining question is how FSD can possibly cope with 'local' problems, which have no established precise address? That is a very common situation globally. Humans stumble seriously on those. I encounter those every time I drive in either Miami or Rio de Janeiro where single addresses serve for multiple buildings with no established logical rules and non existent signage.

Does anyone have qualified opinion?

 

[Bouboule]

For those wondering how big a lead end to end AI is versus other competitors. I have a Rivian and, first, all competitors self driving is highway only. No one is working on urban driving like Tesla has with FSD Beta. Second, with the Rivian, I caught it making a mistake that Tesla had circa 2016. Driving in the left freeway lane, the system can get confused about where the fog line is based on k-rails, repeated posts on top of the k-rails, etc. Using that as a marker, Rivian, at least, is 8 years behind Tesla.

Also Elon said they will be spending $2B this year and another $2B next year on AI hardware compute alone. Not many companies can afford that.

Ya, the only car compagnies that are 3-5 years away are the ones that will adopt FSD .

Smaller OEMs that have no chances of developing autonomous driving might actually move faster to partner with Tesla and give themselves a lead down the road.

It's a huge risk for OEMs if they really do give a 100% market share of autonomous cars for 3-5 years. As FSD gets better, the risks gets much higher and I don't think certain OEMs will stay on the sidelines much longer.

 

[jerry33]

That premise is false. Throughout Europe cars routinely cross national boundaries. In the US most top medical personnel in Detroit cross the border with Canada twice a day, In Texas cross between border with Mexico every day is quite routine. Factually the fundamental issue is how the NN's can emulate a human who is familiar with those idiosyncrasies but with only visual cues.

Does anyone have qualified opinion?

I wasn't clear enough, sorry. I considered E.U. as one country--just as the driving patterns are somewhat different in different cities in the U.S., countries frequently traveled between would be grouped together, but there's no need for a U.S. registered car to know about conditions is the E.U. China, or Australia, etc. I was referring to how many cars move registration from one country to another. Even U.S. to Canada (or reverse) is pretty uncommon though not unknown.

 

[Buckminster]

I'm very glad Elon dropped in here. IMO, this was a lot more informative than the live stream drive.

I wasn't happy with Elon's answer about Hardware 4. It makes me think that FSD beta on HW4 will require an all new training set and thus it will be a long time before Tesla has enough data to get it running. I'm guessing they will prioritize HW3 for the next year or so because they are getting so close to something truly great.

I want my F - S - D!
(on hardware 4)

I hope and expect that the 11.4.7 team move immediately to HW4 V12.

 

[Buckminster]

The X is insane for the UK. My S was way too wide, and got be damage bills to prove it. The biggest fail in the design of the Y is that its not narrower. I would actually have paid MORE to make the car an inch or two narrower. At least its shorter than the S!
Even if the cybertruck is made for our market (doubtful as tesla have openly declared they don't care about RHD now...), it would be absolutely unusable here.

MS = 2189mm
MY = 2129mm (-70)
M3 = 2088mm (-41)

It's an important 70mm.

 

[AudubonB]

It took me some work to find it but I finally did

"Under this agreement, we are obligated to, among other things, meet employment targets as well as specified minimum numbers of personnel in the State of New York and in Buffalo, New York and spend or incur $5.00 billion in combined capital, operational expenses, costs of goods sold and other costs in the State of New York during the 10-year period beginning April 30, 2018. On an annual basis during the initial lease term, as measured on each anniversary of such date, if we fail to meet these specified investment and job creation requirements, then we would be obligated to pay a $41 million “program payment” to the SUNY Foundation for each year that we fail to meet these requirements. Furthermore, if the arrangement is terminated due to a material breach by us, then additional amounts may become payable by us."

from the TSLA 10Q

So they need to keep employment up there until 2028 at least. My thought would be to ramp up a battery cell and/or powerwall/mega pack site.

Not all the lablers would be able to tansition to the new jobs but you could at least hire new people and keep the jobs in total above the minimum.

To be brutal, it appears to me that were Tesla to fall foul of those landmarks in, say, its final years, then paying the penalty of $41MM for each of those years might be considered as a lesser cost to the company. I use the word “might” because it is unclear whether “material breach” also applies.

 

[FSDtester#1]

I'm very glad Elon dropped in here. IMO, this was a lot more informative than the live stream drive.

I wasn't happy with Elon's answer about Hardware 4. It makes me think that FSD beta on HW4 will require an all new training set and thus it will be a long time before Tesla has enough data to get it running. I'm guessing they will prioritize HW3 for the next year or so because they are getting so close to something truly great.

I want my F - S - D!
(on hardware 4)

I was one of the ones in the back seat during the livestream, it was better in the car than you saw on video. Oh, and we did end up finding Zuck. Ashok ended up beating him to a pulp.

 

[Krugerrand]

So what is the licensing fee going to be? What amount will be deemed ‘fair’ by Elon particularly when we consider he’s given Tesla SuperCharging Network away for free to OEMs.

 

[SOULPEDL]

Elon's live stream made me sad because I'm missing out now that I'm on hardware 4.

Remember that slogan, "I want my MTV"?

Well, I want my F - S - D !

A-hum, I believe I'm first in line, (Edit since 2021).

Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

Pre-trading day change-my-view: The whole Robotaxi discussion here is so US-centric it hurts - particularly in this forum where we all should be looking at things from a global perspective. Elon has commented more than once that he thinks Autonomy will happen later in Europe due to...

teslamotorsclub.com

 

[AudubonB]

That premise is false. Throughout Europe cars routinely cross national boundaries. In the US most top medical personnel in Detroit cross the border with Canada twice a day, In Texas cross between border with Mexico every day is quite routine. Factually the fundamental issue is how the NN's can emulate a human who is familiar with those idiosyncrasies but with only visual cues.

Even the concept is difficult to understand for most of us. Decades ago I was involved in state-of-the-art AI, back when they were "expert systems". The advances in Neural Networks have become so rapid and excellent that human-level reasoning is limited by comparison.

Every day I use Google Translate or another; they are now better than any human aides we sued only two years go. Now we look at traffic behavior as yet another topic that seems, for the first time, actually on the cusp of success.

My remaining question is how FSD can possibly cope with 'local' problems, which have no established precise address? That is a very common situation globally. Humans stumble seriously on those. I encounter those every time I drive in either Miami or Rio de Janeiro where single addresses serve for multiple buildings with no established logical rules and non existent signage.

Does anyone have qualified opinion?

I can, perhaps, address your address conundrum. In Japanese urban areas, once you as a taxi driver, hapless newcomer pedestrian, etc. have worked down to the penultimate or antepenultimate “-chome” (city district) portion of an address….you thereafter are clueless. You still have to determine where the “ban-chi”, or city block is - and that is a function of when the block was registered: it is independent of geography. Once you’ve stumbled upon that, you still have the hurdle of the "-go", the house number, which again is not a function of geography but of the time sequence of construction. You cannot expect 48-go to be between 46 and 50, or 47 and 49.

So, what to do? Absent prior knowledge, which could include that your desired ban-chi is on named street X, or that the AI knowledge universe includes a prior vehicle having stopped at the 48-go house, or computerized access to that region’s development data(!), then there is no option to true mapping, down to the level of each structure (which, by the way, at least in Japan, is based on and de facto the same as the aforementioned development plats).

Summary: at the final level, in such urban areas….mapping it is. I wonder how computerized Tokyo, etc., have embodied such platting.

And a shout-out to @hiroshiy for any further insight and especially corrections. But if he’s from Kyoto that’s not operative, as that city has its own system (as does Sapporo, where sanity reigns. It was a blessèd relief for me to live there after Tokyo).

 

[unk45]

I can, perhaps, address your address conundrum. In Japanese urban areas, once you as a taxi driver, hapless newcomer pedestrian, etc. have worked down to the penultimate or antepenultimate “-chome” (city district) portion of an address….you remain clueless. You still have to determine where the “ban-chi”, or city block is - and that is a function of when the block was registered: it is -independent of geography. Once you’ve stumbled upon that, you still have the hurdle of the house number, which is again is not a function of geography but of the time sequence of construction. You cannot expect #8 to be next to 6 and 10,or of 7 and 9.

So, what to do? Absent prior knowledge, which could include that 6-ban is on named street X, or that the AI knowledge universe includes a prior vehicle having stopped at house #3, or computerized access to that region’s development data(!), then there is no option to true mapping, down to the level of each structure (which, by the way, at least in Japan, is based on and de facto the same as the aforementioned development plats).

Summary: at the final level, in such urban areas….mapping it is. I wonder how computerized Tokyo, etc., have embodied such platting.

And a shout-out to @hiroshiy for any further insight and especially corrections. But if he’s from Kyoto that’s not operative, as that city has its own system (as does Sapporo, where sanity reigns. It was a blessèd relief for me to live there after Tokyo).

Perhaps one of my least known characteristics is that I learned to drive in Tokyo and a small town in Northern Honshu as well. Back then there were attendants/policemen at every small intersection whose primary function was to advise people where a given building was. Today there are a plethora of Japanese localized mapping devices and even some in English:
:‎Tokyo Map
As you suggest there are some situations in which no logical solution exists so physical mapping is the only probable solution.