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

[Chunky Jr.]

Tesla settles CA lawsuit over handling of hazardous material...

The lawsuit claimed that Tesla improperly labeled waste such as diesel fuel, paint materials, lubricating oils, brake fluids and used lead-acid batteries, among others, sending the materials to landfills that don’t accept these types of waste. The suit also names nine violations related to the handling of waste, from disposal and overall handling to transportation and employee training.

While Tesla did not admit to doing anything wrong in response to the suit, the company agreed to the settlement amount and said it will take steps to better handle waste in the future, along with hiring a third-party auditor to monitor its waste handling over a five-year period. The automaker also said it had already begun screening its waste more carefully.

Tesla settles hazardous waste lawsuit in California for $1.5 million

Tesla has agreed to pay $1.5 million to settle lawsuits alleging it mishandled and mislabeled hazardous waste materials.

www.teslarati.com

 

[Knightshade]

While Tesla did not admit to doing anything wrong in response to the suit, the company agreed to the settlement amount

We will never surrender/settle an unjust case against us, even if we will probably lose.

 

[lascavarian]

I disagree that the experience might be worsened in one of the places. The samples from both cities might include local information that makes the driving better in both places, using the local information.

Thanks for your comment. I wish you were fully correct but I am not so certain. It could result that a single example in the training set improved many experiences in many cities but similarly the next example might regress half those cities.

I don’t understand their training method to be very susceptible to easy curation for specific results. This would swap the hard problem of hand coding for the hard problem of hand curating.

Rather the progress comes from both some curating coupled with huge amounts of new data. If this different training data improves the result then great. It might not in which case the curating and building continues.

I expect there are clever wrinkles also happening so I would love to be surprised. The one thing I will pay attention to is the rate of iteration for V12 vs V11. I think V12 will be a lot slower as to the release schedule but just a guess.

Again thanks for your comments

 

[Andy O]

This would swap the hard problem of hand coding for the hard problem of hand curating.

Thank you for the reply. I agree with you.

I wonder if the input date to the FSD trainer includes the location of the driving being analyzed, and a geopolitical map of N. America. In this case the output code could cause behaviour that varies depending on location.

 

[Skryll]

Well if you haven't noticed, immigration is a huge concern for the majority of the country, it's currently the #1 concern among voters, 81% of Republicans, 68% of independents and 45% of Democrats say conditions at the border are worsening. So Elon is not expressing some controversial view here, his tweets are in broad agreement with the sentiment of most Americans.

Because of misinformation campaigns like his.

 

[growler23]

At this point, I think it goes way beyond artificially depressing the SP to buy shares on the cheap. Indeed, I don’t think it’s that at all. I won’t bore people with my conspiracy theory.

My father is really, really, really considering his first EV. It’ll be the last vehicle he buys in this lifetime. As a mechanic his whole life, he’s been having a real hard time undoing decades of ICE knowledge and expertise. I keep having to tell him, ‘Dad, it doesn’t work that way with a Tesla.’ (Absolutely I will not allow him to buy anything but a Tesla.)

As I eluded to recently, I had a several hour conversation (read spirited debate)with him the other day about many things Tesla related that he’d learned by listening/reading news/media and the like.

He’s not a gullible man. He’s not slow witted. He has a healthy amount of skepticism regarding the media, but even with all that I had to set him straight on a few matters.

Now the typical reaction to that is ‘Tesla, needs to do this or that to educate people, rebut the media, yadda, yadda, yadda.’

Meh. It wouldn’t make a significant difference overall. Linking people to Tesla produced information (like their blog, impact reports etc) doesn’t suddenly make people believe. Indeed, they can often think the company is lying because company’s do often lie, so for them ‘what makes Tesla different in that regard?’. In many cases an ‘independent’ source is more believable to them.

*This* has to run its course. The good, the bad, and the ugly of it all. People as a whole have to arrive at the right answer on their own, in their own way, or it means nothing will change, nothing can change for the better of mankind.

I feel despair for Elon, his companies, all the employees, all who are working so very hard to help us all and who are constantly being spit upon. It’s classic good vs evil. I believe the good win this time, though, not without a lot of casualties along the way.

The Fossil fuel industry is so old, profitable, powerful, well entrenched, highly mouthpieced, politically and legally protected and funded, with its tentacles well insinuated in our assumptions and thoughts from years of ads, that word of mouth from a trusted source may be the only way past it, and even then we hit roadblocks.
I reiterate: Civilization wide, this Transition will not be an easy choice on a sunny day; it is more akin to ripping out our own guts.
Tangential observation: Areas of the world that are currently in conflict and areas that are massively tied to Fossils for their power... seem to overlap at lot of late. USA included in some ways, 'nuff said there.
To all: please keep spreading the facts when you can. Tesla will keep makin' the goods. Last year, we added 1.8 million possible story sharers.
Next year, imo, Gen3 will cut the head off the already mortally wounded dinosaur.

 

[Cult Member]

I am beside myself with excitement! Rumor has it that the Cybertruck is coming to Germany, especially to Stuttgart (Tesla delivery center Holzgerlingen, very close by). Unfortunately (for now) only as an exhibit, but still. A great start.

If the rumor turns out to be true, I'll pitch my tent in front of the showroom and be there every day (if my vacation request is approved).

By the way, on the thread topic: Last week I was able to get my stock to my target size, a nice round number. Liquidity therefore close to zero. Now the share price can rise again. Many thanks to the (non-)advisors here in the forum! Hurrah!

https://twitter.com/x/status/1754124475778359471

Congratulations to your target size!

Would you say that April, 13th is a possible date for a Cybertruck presentation? Timo Schadt made a vague save-the-date announcement.

 

[AudubonB]

I am beside myself with excitement!...Last week I was able to get my stock to my target size, a nice round number. Liquidity therefore close to zero. Now the share price can rise again. Many thanks to the (non-)advisors here in the forum! Hurrah!

100,000,000 shares? Would you care to come over for dinner?

 

[cab]

That's not the way it works. Learning is cumulative. Just because FSD is better in San Francisco does not make it worse somewhere else.

FSD V12 needs more training in rainy areas because it needs more training in rainy areas. Once that happens, V12 will get better at driving in the rain. But that won't make it worse in San Francisco.

If you oversample in a geofenced area, the FSD will get really good at driving in that area. But that won't make it worse anywhere else. And while it is getting good at driving in the geofenced area it will still get better driving other places as more and more training data is used from those other places.

I think we need to understand that geofencing is the only way to get started with a robotaxi service. There are lots of reasons that should be obvious. If you want, we can do a longer discussion.

Eventually, FSD will be good enough everywhere that you can start a robotaxi service anywhere. But that day is far from now and the logistics of such a universal service can not be put into place quickly. It will need to grow organically over a longish period of time.

Or think of this. Uber started out geofenced and it still is as it is not available everywhere. But at the start, it was only available in certain cities. And Uber didn't have to solve FSD. It was geofenced because of logistics. The same will be true when Tesla starts its robotaxi service. You have to start somewhere, not everywhere.

Agree with "geofencing" in a specific area to start with for robotaxis. Geofencing, however, isn't entirely the right term for this as compared to say what Waymo is doing. I think the latter (and Cruise, and tech like GM's Supercruise), rely on lidar maps of the area in question. In the case of FSD v12+, it really just comes down to a high enough confidence level in the software FOR A SPECIFIC GEOGRAPHIC AREA. This may mean you can use FSD in San Francisco (or the Bay area, etc.) and a few other cities initially. The use case for traditional Taxis and Ubers is likely within metropolitan areas so having the software work in those areas delivers immediate value, while other areas are still being flushed out. I just don't think it's necessary to say "every conceivable scenario in EVERY area across the country must be solved before ANY area can benefit!". Indeed, an OTA update that enables "Los Angeles or SoCal for FSD" is a heckuva lot more impressive than fart sounds!

 

[lascavarian]

I wonder if the input date to the FSD trainer includes the location of the driving being analyzed, and a geopolitical map of N. America.

This is a great question. I “think” their design model would try to avoid this but on some level it might come into play.

 

[hobbes]

Some fun to end the weekend..

 

[woof]

I may not need that pitbull afterall.

So Goldie is admitting dogs are useful. But why she thinks a cuddly cute puppy would keep people AWAY from her truck is baffling.

 

[Musskiah]

For those who are into TA (or T&A)

 

[Singuy]

Guess Telsa is expanding advertisement. You now see youtube ads

 

[Usain]

To be clear I don't think it's an unsolvable problem, but I DO think it means overfitting to one area CAN make it worse in others.

I need to be more clear as well. I am assuming that Tesla engineers know what they are doing. In the case of Tesla FSD, overfitting will make it better in a geofenced area without making it worse somewhere else.

And as you said, the problem might be hard, but it is solvable with end-to-end. For your turn on red example, the training data includes the location of the clip it is learning from. The system will learn not to turn right on red in places where that is against the law. If there are signs, it will learn from the signs. If the law is unclear, it will do no worse than a human.

 

[Usain]

Regarding Right on Red there's an intersection on75th & Rt64 in Bradenton where a white No Turn on Red LED sign pops up often & under control of crossing guards. It's way up high and nowhere near the traffic lights. Stuff like this and other drivers waving to you indicating you may proceed as I see you will make L5 and Robotaxi revenue $ a LONG way away.
Back to stock issues.

This is exactly the kind of thing end-to-end will solve. It's really hard to teach about drivers waving to you if you want to teach that heuristically. But the end-to-end system will pick that up once it has been trained on enough clips of "waving" behavior.

 

[jerry33]

This is exactly the kind of thing end-to-end will solve. It's really hard to teach about drivers waving to you if you want to teach that heuristically. But the end-to-end system will pick that up once it has been trained on enough clips of "waving" behavior.

So I guess the car should respond when it sees a rude sign directed at it

 

[Usain]

Kind of agree. In Winnipeg the lights are always red, but green arrows (or lights) show in addition to the red lights to indicate when you can go (not just for turn, but for straight ahead as well. I don't know of any other city that does it this way. (It's been years since I've been in Winnipeg, so it's possible it's been changed.)

Yeah, this is why overfitting is the way to go. If you are starting your robotaxi service in Winnipeg then you train the system on lots of extra Winnipeg clips and it will learn to do the right thing. If you don't do the extra training, then the system should do no worse than a human visiting Winnipeg for the first time.

But overfitting for Winnipeg would not make the overall system worse because those special "Winnipeg style" traffic lights usually won't show up in other places. If another city does do it that way it will treat it as a Winnipeg light. If a "Winnipeg style" light means something different in another city then the system will do no worse than a human visiting that city from Winnipeg.

 

[Usain]

Nothing in this process can prevent regression. Over sample in SF will change the FSD experience in St Louis for example. Every time training data evolves results evolve.

If you are right then we all need to sell our stock.

Learning to drive better in one city should not make you an appreciably worse driver in another city. It's just that you've picked up the subtlties of driving in that first city. Learning is cumulative.

Now granted, we are taling about artificial neural nets and not biological ones. So the comparison is not perfect.

But again, if Tesla's engineers are not good enough to deal with regression we should sell our stock.

 

[Usain]

Thanks for your comment. I wish you were fully correct but I am not so certain. It could result that a single example in the training set improved many experiences in many cities but similarly the next example might regress half those cities.

I don’t understand their training method to be very susceptible to easy curation for specific results. This would swap the hard problem of hand coding for the hard problem of hand curating.

Rather the progress comes from both some curating coupled with huge amounts of new data. If this different training data improves the result then great. It might not in which case the curating and building continues.

I expect there are clever wrinkles also happening so I would love to be surprised. The one thing I will pay attention to is the rate of iteration for V12 vs V11. I think V12 will be a lot slower as to the release schedule but just a guess.

Again thanks for your comments

I think the improvement rate for V12 will accelerate and become much faster than any release before it. I disagree, but I gave your post a like because I like it.