Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

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

This site may earn commission on affiliate links.
A new startup helm.ai uses unsupervised learning to teach deep neural nets to do autonomous driving, any expert can comment if this could be a competitive threat to Tesla as it might neutralize Tesla's fleet size advantage? See Forbes article: Forbes Insights: The Future Of AI Is Unsupervised where the CEO talks about no need for labeling : "With unsupervised learning, you feed the model images without labels and it learns to understand those images purely algorithmically. Removing the need for labeling means you can train models faster and more cheaply."

You still need to feed the data (images, videos, LiDAR scans, whatever) into the system for it to unsupervised-edly label. If you don't have a fleet to speak of, you'll be severely restricted as to the amount of data have have available to feed it. (Whereas Tesla will always have mountains of data to feed their training systems, whether they used supervised labeling or not.) Removing the bottleneck of labeling speed doesn't do anything about the bottleneck of available data to label.

So no, I don't think this would in any way neutralize Tesla's fleet size advantage.
 
I was listening Dan Carlin's "Wrath of Khans" talking about "creative arsonist" clearing up the "deadwood" to create space for the later civilization, and found similar parallelism of today's world -- there are lots of industrial deadwood up for grab and COVID has accelerated the process.

It mentioned about how Mongol could conquer the enemies (fossil fuel industry, energy sector, aerospace, telecom and pharma) with a fraction of its force (OPEX). The force that SpaceX, Tesla, Boring and Neuralink methodically decimate the traditional old guard at full front.

If you worry about Elon's tweet how might it impact short term stock price, please do us a favor, sell it now -- and invest in other companies that makes you feel less sinful. Elon doesn't not pretend his nobel class like other hashtag, pronounce CEOs, and that does not bring his army less fearsome.

If I were one of the deadwood, I would shake in my boots. Just wait until CureVac vaccine goes Phase 3 early next year, a lot of Elon COVID bashers would turn their head 180 degrees and hail the Tesla's Bioreactor as one of the medical revolution that brings vaccine into scale where others are still trying to figure out storage and transportation.

If Elon wants to disrupt something, he could do it in a blink of eyes. The pharma complex has pissed him off. Just wait him to bring his wrath.
 
Is seems to me that people don’t understand that FSD is not just a piece of software that as soon as a company gets to solve it, it will just work on any machine?

They will need the machine/s to be able to run the FSD, not just the software. It’s the machine with the software that is the complete package.

What’s up with this concern that “ omg, someone else will solve full self driving and TSLA will drop!!!!” ?

People don’t seem to get that tradition oems can’t even do a freaking OTA, but somehow overnight they will make machines to drive themselves?

4D chess is not checkers, is it?

I liked what Elon said in regards to batteries:” we are the biggest buyers of batteries, where do you think is the 1st place that someone will go if they have a better battery?”

if someone will truly solve FSD before Tesla, will come to the only car manufacturer that has a car on the road NOW that can run that. And the only manufacturer that has 3 ginormous factories getting built, the only manufacturer that has the best battery technology that needs to be scaled... the list goes on.

I might be wrong, so please school me! :) TIA
 
We don't need to know the actual cost to know it will be dramatically cheaper.

We need to know the actual cost to judge it it'll be SO MOAR CHEAPER people stop owning cars though.

We know at current pricing of rideshare and taxis that people still prefer, in massively massively, overwhelming numbers, to own cars.

If you want to make the argument there's some lower cost that is SO low a large % of those people will change their mind on that, you'd need to:

Know what that lower cost is (not sure how you determine this)
and
Know what the actual real world cost of an RT would be, to compare against item 1.


As I pointed out, Uber nationally runs about $2 a mile. So I'm not sure "50% off uber" is cheap enough to get folks, especially Americans, to ditch car ownership in numbers as large as some folks seem to think will happen.

Certainly not in numbers to support throwing 10x more RTs on the road today than there are uber drivers... (and that'd be ONE year of Tesla car production by 2030 or sooner per Elons production estimates).


Come on... electric, low maintenance, zero driver-labor, zero-marginal-cost software...

zero to whom?

The discussion was originally around individual owners running their cars as RTs... so the software is nowhere NEAR 0 cost to them.

If you're switching to Tesla running their own fleet without anyone else involved costs obviously drop- but that promise to owners they can make a ton of $ on their RTS take a kick in the groin since Tesla can do it a lot cheaper than individual owners can...like a LOT cheaper... they don't need to lose 30% of fares to themselves as individuals on the Tesla network would... they don't need to make 20+% gross profit on the vehicle sales, the profit on the software, and can do charging and any maintenance a bit cheaper than individuals too.

So that's really 2 different discussions.

Tesla owned RTs will have massively lower cost than individual ones. And know they're gonna have a big fleet of em based on the lease stuff going on.




Here is my argument in a nutshell, which you have yet to refute: You keep estimating robotaxi usage based on cheap transport options that exist now, but there has never been a cheap option with the comfort, convenience, privacy and safety of Tesla robotaxis at scale. New technology can change behavior.

But this isn't "new" technology to the user.

"Car I can hail with my phone that picks me up and takes me somewhere safety for money"

That's Uber, but cheaper, and without annoying conversation.

Certainly that's BETTER than more expensive and WITH annoying conversation.... but I can't think of a time I ever WOULD have used Uber, but didn't because it wasn't just a few dollars cheaper, or because I didn't want someone asking me how my day was on the ride.


I've certainly never thought "Man, if Uber was 50% cheaper we'd just not even own any car at all"


So my argument, which you've yet to refute, is that RTs when/if they end up existing outside of Chandler, AZ, will replace nearly 100% of "regular" taxi and human-driven rideshare vehicles.

Which is enough to absorb about 10% of 1 years 2030 Tesla vehicle production.

Probably they'll see a significant increase in use from folks who no longer bother repairing/keeping an older 2nd or 3rd car around for only occasional use.... so MAYBE that doubles the # of rideshare vehicles needed? Given utilization stats that may even be high, but let's go with it.

That's 20% of 1 year of 2030 Tesla vehicle production.

So again the idea anybody can just make $30,000/yr on their car in their spare time as an RT seems largely fantasy in the long term.

And again Tesla THEMSELVES will be running a fleet of these- cheaper than individual owners can do so.



And actually, I did ride the NYC subway. As I recall, there was lotsa noise, dirt, hard or unavailable seats, and no way to talk during rush without six strangers eavesdropping. That's what I meant about privacy.


Which spy agency do you work for, specifically, that random subway rider knowing what you plan to eat for dinner is a big problem?

Mostly New Yorkers avoid eye contact and give not a crap what anybody else is saying, FYI.

You DO occasionally have to stand during busy times (well, pre covid anyway).... but you get there in 5-10 minutes instead of sitting in traffic at 3 mph for 45 minutes, so that's still a win.
 
  • Disagree
Reactions: StealthP3D
It's a big number, but throwing money at a problem doesn't mean it's either solvable or quick. I'd expect most of it to be wasted.


Yup.

See also VW saying they're gonna hire 10,000 people for a new software division to get software right.

As has been cited, they think if 1 woman can give birth in 9 months, 9 women can do it in 1.
 
It's a big number, but throwing money at a problem doesn't mean it's either solvable or quick. I'd expect most of it to be wasted.
It is a big number. Still, it is approaching the order of magnitude, ~$100B, that I’ve said in times past here would be required for a legacy OEM to get into the EV game for real.

To be sure, most of it will be wasted. That is the nature of innovation in large organizations. I’m sure VW management is aware of this.

At least three things now work in the legacy OEM’s favor if they would get serious about EV’s:

* They have a template in Tesla. Whether they admit it or not, they are incapable of meaningful new, comprehensive and coherent innovation. If they ape Tesla as best they can, they might find a way through the woods.

* There is broad acceptance, albeit grudging, that EV’s are the path forward. Most of the debate within the legacies will go by wayside (at least at those that would survive). The relevant governments will be increasingly interested to help and less prone to pressure from fossil fuel interests (except in the US where I fret any meaningful new federal level assistance will peter out or be derailed, I’m sorry to say).

* Nothing like survival to focus the mind. Executives will increasingly be held accountable and not be allowed to kick the can down the road wrt to EV’s at legacies where they can be held accountable (which, granted, is likely and unfortunately very few)
 
A new startup helm.ai uses unsupervised learning to teach deep neural nets to do autonomous driving, any expert can comment if this could be a competitive threat to Tesla as it might neutralize Tesla's fleet size advantage? See Forbes article: Forbes Insights: The Future Of AI Is Unsupervised where the CEO talks about no need for labeling : "With unsupervised learning, you feed the model images without labels and it learns to understand those images purely algorithmically. Removing the need for labeling means you can train models faster and more cheaply."
Although many seem adamant that Tesla is not using unsupervised neural nets because Tesla isn’t saying so, it is not out of the realm of possibility that they do.

Models such as Adaptive resonance theory - Wikipedia (ART) were explicitly designed to explain how humans and some other animals learn in the real world. Further, categorical learning (supervised or not) can be combined with reinforcement learning that is timing sensitive, e.g. Neural dynamics of adaptive timing and temporal discrimination during associative learning - ScienceDirect

A talented modeler can mix and match various models in non-obvious ways to get striking results. This might be considered the very recipe of a trade secret.
 
I think we should applaud and support VW for (at least trying to) putting their money where their mouth is. Be it $40BN or $90BN, doesn't matter. They are doing something to transition their products to sustainability. I think that's great and do not see it as a threat to Tesla. It's going to take a couple more automotive giants to make this change in order for the world to transition to 100% sustainable transportation.
Good on them.
Agreed, but it's one thing to announce your investment (in this case several times), but another thing completely to do it... successfully at that. ID.3 anyone?
 
You still need to feed the data (images, videos, LiDAR scans, whatever) into the system for it to unsupervised-edly label. If you don't have a fleet to speak of, you'll be severely restricted as to the amount of data have have available to feed it. (Whereas Tesla will always have mountains of data to feed their training systems, whether they used supervised labeling or not.) Removing the bottleneck of labeling speed doesn't do anything about the bottleneck of available data to label.

So no, I don't think this would in any way neutralize Tesla's fleet size advantage.
True enough, though it may unlock a niche application or three within which they might prove their chops and from whence they might grow.

edit: Strictly speaking the categories can be associated with an internal representation of something else —an object, a sound or word, a behavioral response, ... — which are not ‘labels’ per se.
 
Last edited:
  • Informative
Reactions: jamaatta
I assume this is good news for Giga Shanghai demand:
Subscribe to read | Financial Times

The RCEP takes most of the existing agreements signed by the 10 members of the Association of Southeast Asian Nations — Brunei, Cambodia, Indonesia, Laos, Malaysia, Myanmar, the Philippines, Singapore, Thailand and Vietnam — and combines them into a single multilateral pact with Australia, China, Japan, New Zealand and South Korea.

[...]

I also hope this will be good for exports from Tesla's Shanghai factory to nearby countries. However I haven't seen any specific mention of finished automobiles in the RCEP coverage to date — only auto parts.

Has anyone seen specifics about how RCEP will affect tariffs on finished autos, especially BEVs? Perhaps @jbcarioca has expertise here?
 
A new startup helm.ai uses unsupervised learning to teach deep neural nets to do autonomous driving, any expert can comment if this could be a competitive threat to Tesla as it might neutralize Tesla's fleet size advantage? See Forbes article: Forbes Insights: The Future Of AI Is Unsupervised where the CEO talks about no need for labeling : "With unsupervised learning, you feed the model images without labels and it learns to understand those images purely algorithmically. Removing the need for labeling means you can train models faster and more cheaply."
Tesla is using supervised NN today, but they have never said they will continue to supervise forever and ever and ever, have they? For all we know, they may Already have plans for switching to unsupervised networks when they have the confidence that the current process works great.

As any good manager knows, when you are training humans new at their job, you first supervise still they become good at their job. Once they are good then you think of switching to unsupervised. If a manager expects to Start with unsupervised work ethic they’re setting up the new person for failure. On the other hand if a manager plans to continue supervising everything forever he’s a really bad manager. I see neural network and AI exactly the same way.

Disclaimer: I understand nothing about artificial intelligence or neural networks. But I do manage large teams of people.
 
I sincerely hope STMP does not make a habit of taking others’ insights, repackaging them with an occasional “That’s a good point!”, and asking people to contribute to his own kitty. There are other ways to Solve Your Money Problem.

As long as it’s once in a very great while, and you bring your own insight to the table, then I’ll not criticize again. It’s always a pleasure to re-listen to Peter Lynch.

BUT...if anyone defends that amazing plagiarism then you've no right to expect to be taken seriously if you criticize other kinds of broadcast dishonesty, as in a lot of what appears on CNBC. Capisci?
 
I assume this is good news for Giga Shanghai demand:
Subscribe to read | Financial Times

The RCEP takes most of the existing agreements signed by the 10 members of the Association of Southeast Asian Nations — Brunei, Cambodia, Indonesia, Laos, Malaysia, Myanmar, the Philippines, Singapore, Thailand and Vietnam — and combines them into a single multilateral pact with Australia, China, Japan, New Zealand and South Korea.

By combining a mishmash of separate arrangements into a single deal, RCEP brings Asia a step closer to becoming a coherent trading zone like the EU or North America, even if it is not expected to lead to large overall tariff reductions.

China’s premier Li Keqiang described the agreement as “a victory of multilateralism and free trade”, according to a report from the official Xinhua news agency.

Japan and South Korea are expected to be among the biggest winners from the deal, but the benefit of cheaper goods will spread as far as Europe and the US.

“We signed [RCEP] today, after a tough slog of eight years,” said Lee Hsien Loong, Singapore’s prime minister, on Sunday. “This is a major step forward for our region. At a time when multilateralism is losing ground, and global growth is slowing, the RCEP shows Asian countries’ support for open and connected supply chains, freer trade and closer interdependence.”

Australia and China are in the midst of a trade war. Does this settle all of those issues? Any Aussies care to comment?
 
Last edited: