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.
Do you have any idea how ridiculous you sound? The people who can most accurately comment on Tesla's FSD progress are those testing it in the real world. There are lots of those people. Many are unqualified to offer any kind of opinion because they understand little about how to accurately observe anything. But that still leaves plenty of people with direct experience and the ability to interpret and describe it.

The Tesla team can most accurately assess the progress of those things that haven't yet made it out to their legion of beta testers.

Personally I have observed only marginal improvements since last year, when I first got the FSD Beta software. They have come slowly and steadily, but with many steps backward. And I'm mostly driving around Bay Area suburbia, which ought to be pretty easy and familiar to Tesla's developers. There have been no "step changes", which I've found very disappointing.

I don't think I expressed my point well.

Yes observers see and measure progress to some extent, but the apparent lack of progress doesn't really inform us about the strategy relative to other strategies, future progess, or why particular regressions are happening.

My point was mostly aimed at those suggesting Tesla should be doing things differently.
 
Good thing this didn't crash TSLA. Keywords: Tesla fire car crash.


Absolutely nothing to do with Tesla, Inc.
"Corral Hollow" Fire wouldn't get anywhere near the clicks.
 
Could I ask an ”uneducated” question please. The assumption seems to be that if enough data is thrown at a NN, it will eventual learn to do the right thing. Has this ever been proven for a moderate task? Ever since my first day as a programmer (IBM mainframe, punch card area), AI has promised to obsolete the programmer, and 40 years later the software guys are as necessary as ever!

The example often cited is AlphaGo where the machine can beat all human players.

FSD is a hard problem for traditional programming, humans coding for all edge cases is daunting.

My view is the chances of Elon and Karpathy both being wrong is small, and I am not aware of an alternative solution that seems to be making progess.

I am aware that FSD goes though multiple versions, where the approach needs to be revised. I see it as FSD and the team slowly learning from mistakes and from things that they got right. The timetable slipping is becuase the problem is hard, but it is equally hard for everyone.

The team has a lot of hard numbers on how they are progressing, it seems that occasionally those numbers might cause Elon to be overly optimistic.
 
What's wrong with Elon? He hasn't tweeted in a few days
According to ElonJet he's just been in Austin for the last couple of weeks (with the exception of 1 day in Brownsville). Perhaps he's gone into factory mode to help get Austin ramped (i.e. being seen by the new employees on the factory floor helping to get the new lines sorted, etc).

Or maybe he just hasn't seen any good memes lately.
 
You don't think that workers have the right to union?
Tesla has repeatedly allowed their workers the opportunity to unionize...but is also clear that their compensation plan will change and they will lose their stock options in order to do so. It's financially in their best interest to NOT unionized solely from this aspect of their current compensation. Long runs of time with poor stock price performance makes Tesla more vulnerable to the unions making inroads....as long as the stock continues to appreciate in a fashion that makes options lucrative, the unions shouldn't have a play.
 
Could I ask an ”uneducated” question please. The assumption seems to be that if enough data is thrown at a NN, it will eventual learn to do the right thing. Has this ever been proven for a moderate task? Ever since my first day as a programmer (IBM mainframe, punch card area), AI has promised to obsolete the programmer, and 40 years later the software guys are as necessary as ever!
Andrej Karpathy likes to quote Ilya Sutskever:
With a large enough network, a large enough labelled training set, and a large enough budget for GPUs, success is guaranteed

1656229139564.png


For the rest of the slides where this is explained:

Basically, if you have a large enough neural network, train it for long enough on a large enough dataset(that is well labelled), it will learn EVERYTHING that can be learnt from the dataset. So if it is technically possible to do it somehome, ie if a super intelligent human could figure it out given infinite amount of time and the entire dataset, the neural will also do it.

Some problems such as “here is 8x8 chessboard, you have one forward pass to figure out the best possible move” might benefit from some form of search technique, but a large enough neural network, with a large enough dataset and long enough training will in theory just be able to solve it without using any search. If we can fit the neural network in memory using all the atoms in the universe is another question. And yeah, fitting the labelled dataset in our universe will probably stop us before we even try to do it that. But long before we need to get there we will have a solution that is pretty damned close to drawing from the starting position and winning any winning position if the opponent ever makes a losing mistake.
 
Last edited:
FSD beta provides a very good value when you are in a completely new city/area. I was using it in Tampa and had zero clue where I was going. FSD beta just took over and I just let it do it's thing. NOA +FSD beta on our way back to central FL completely bypassed a hour worth of traffic by taking an exit unexpectedly and drove around on city streets while the hwy was jammed due to construction. I had zero clue where the car was going and it was pitched dark in these country roads. I just let it do its thing and I ended up by passing 6 miles worth of bumper to bumper traffic as it drove me back onto the hwy with near zero interventions. My wife afterwards was like "wow it did a good job this time, good job Tesla!".

I own a 2013 non-AP P85+ And Google Navigation does this for me, not NOA/FSD/AP
 
I own a 2013 non-AP P85+ And Google Navigation does this for me, not NOA/FSD/AP

No, it does not.
It just suggest that YOU should do the turn when you did not expect it, so you didn't pay attention and missed that exit.
This gets supper annoying on regular venues where I do not need navigation to lead me except for detours due to traffic etc.

If the car navigates, I do not need to pay attention to navigation suggestions every single day and minute ...
 
Yeah, an integrated PowerWall V2G capability would make sense.

The other functionality could be DC-DC style fast car charging... i.e. "Supercharger Lite" ... using the PowerWall pack...
Volkswagen is pushing/lobbying for V2G via DC for a while now.
They came to the same realization that AC connection doesn‘t make any sense as efficiencies are too low and cost overhead in the car too high.

They started presenting a DC/DC v2g Wallbox prototype in 2020 and are testing since then:

Pilotphase für innovative DC-Wallbox gestartet

1656241714454.jpeg


My hope is that Tesla, being present in Germany now, is starting to show a concrete vision on where to move current regulations to - Volkswagen is doing exactly this for a few years now and lobbying towards regulations that make it possible.


My personal opinion is that we need every battery on the grid as much as possible and need incentives for car owners to make a certain percentage of their battery available to grid services.

Tesla just started the VPP project with PowerWalls - and tbh, this makes great sense to start with 100% Available batteries first and integrate the more difficult option of car-batteries later.
 
A company with a union ceases to be a nimble company.

A union in a well functioning company like Tesla is bad for the ambitious workers, bad for the company, bad for the stock. It is only good for the union leaders.

ROFL @9837264723849 disagreeing with this, the self-proclaimed French Socialist, there is a shock (not).

Remind me what the French are known for in industrial prowess? Can you even be good at high-volume production with a 35h work week and unionized labor?



The the point of @mars_or_bust, Tesla requires that people be FLEXIBLE in their work description, because immediate needs may change (i.e. a buddy of mine, a VP, was called in to help on the line during 2018 for the Model 3 end-of-quarter push). Unionized positions specifically AVOID that kind of flexibility. Numerous posters here have shared their union experience with traditional auto where a piece of equipment on the line fails, or there is some other problem, and all work stops because everyone says "I know exactly what is wrong, I know how to fix it, but I'm not responsible for that piece of kit, so I will not touch it". Waiting hours, or sometimes days, for the proper union person to show up and perform the repair, so everyone else can get back to work.

There are NO EXAMPLE of unionized companies being nimble. And to survive and excel in today's business landscape, those two things are required.
 
Anyway workers have the right to union if they want.

And the owner of the company has every right to fight against it. Let's not forget that the shareholders have rights here, and Elon is the largest of those. People want to gloss over that fact, the fact that the rights of the workers are NOT the only "rights" under question.

Again, it's not like Tesla mis-treats their employees. They have THE HIGHEST pay rate among US-based auto companies, and that's before accounting for stock options, which I believe they are the only auto company to provide.
 
I'm not a fan of unions, but here's an alternative view:


And my counter-argument to that would be this:

Those "Made in USA" manufacturers have over the past 30 years outsourced more and more of auto and auto-component production abroad. Where? To countries for the most part which unionization pretty much never happens (SE Asia, Mexico, etc.). This has gutted the majority of the jobs for auto manufacturing in the USA, to a level far far lower than they would have been if there was no unionization but still good and proper compensation for workers.

Tesla, the opposite. It now has 4 of the 6 top "Made in America" autos:

Furthermore, there is INSANE corruption at the UAW, and it appears to be at all levels. By any reasonable outside observer, it appears that the main focus of the UAW is more on the enrichment of the UAW management and upper tiers, than on "rights" for their workers. Rights, I will add, that are already guaranteed strongly by various laws. We are not back in the late 1800s in sweat shops over here.

And then there is the insane spending on political lobbying but the UAW, which is a large, LARGE part of their budget (not directly on workers' benefits):

I mean, let's be honest, if there were even a FRACTION of the his happening at Tesla, the media would have already set Tesla on FIRE.
 
A company with a union ceases to be a nimble company.

A union in a well functioning company like Tesla is bad for the ambitious workers, bad for the company, bad for the stock. It is only good for the union leaders.
I agree with this statement for the US. For the rest of the world it really depends on the country and situation.
 
I agree with this statement for the US. For the rest of the world it really depends on the country and situation.

Absolutely, Unionization has to be viewed in the wider context of laws that govern and protect those workers. In the USA, those laws have continued to strengthen over 100+ years.

Totally different situation than say . . . Thailand or Vietnam.