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

[Drezil]

Yahoo just now in the premarket:

These are the first "editor picked" news-articles you will see under the autoplaying videos..
actual news (kaparthy leaving, kansas panasonic plant, etc.) are 1 page further scrolling down ..

Nothing to see here. No agenda .. move along.


Edit: so much FUD. Bullish!

 

[Bladerskb]

This WIRED article is comedy gold. Without mentioning Tesla even once adds another subtle layer of irony: the author is probably as confused as those stranded bolts as she’s wondering why the supposedly superior Cruise is so primitive in real world.

https://twitter.com/x/status/1547437065733808128

But Its not remote controlled... stop spreading misinformation

 

[Mengy]

But Its not remote controlled... stop spreading misinformation

It's partially correct, they have remote control backup. Which means they need people manning the emergency backup controls for every single Cruise car on the road. Not a very scalable autonomous solution.

 

[Usain]

Tesla has top notch AI software engineers and I'm sure they will be able to attract more.

So the loss of Karpathy doesn't mean a whole lot at this point. FSD will continue to make slow, steady progress.

It's the "slow" part that bothers us investors. And the bottleneck for speed is just raw compute power. So I think the development of Dojo is far more important than the development of the kinds of ideas that Karpathy brings to the table.

 

[traxila]

“Meanwhile, commodity markets remained under pressure on rising worries of a supply crunch. West Texas Intermediate (WTI) crude futures fell by $2.24, or 2.33% to $94.06 per barrel in the early trade, and Brent Crude Oil fell by $1.94, or 1.95%, to $97.63.”

From our good friends at Yahoo. Journalism just gets better and better.

Has their actually been a supply crunch anywhere for OIL? Rather an inability for certain parties to buy based on prices pushed by speculators.

Also, what the hell are they talking about when they write that if “Russia withholds their oil, prices will go to 300”? What does that mean? Russia will stockpile their 8 millions barrels a day indefinitely, as if they are not desperate for cash? Or that they won’t sell to Western countries, but rather to other parties in Asia…. Which means NOTHiNG as they are selling their oil on the global market and that is all that matters.

Think I might crush all my screens and take a break from the world. Nothing but nonsense everywhere. Not even lies anymore, just incoherence. Our ability to communicate is under assault and degenerating across the board.

 

[heltok]

I have talked a bit with Karpathy, very down to earth over the top nice guy. Don’t know him too well but will just paint a story based on public posts by him and what everyone knows.

He is a pretty young at his 35 years, born in Slovakia. Brilliant guy. Got a Phd(with all that means) at Stanford, at the time when deep learning took off. Being a very articulate he got the role to lead the course that many students around the world watched on youtube. He read so many papers he even did his own service to follow new papers more efficiently.

GitHub - karpathy/arxiv-sanity-preserver: Web interface for browsing, search and filtering recent arxiv submissions

Web interface for browsing, search and filtering recent arxiv submissions - karpathy/arxiv-sanity-preserver

github.com

After his Phd, he got a job at OpenAI as a scientist. One year later he joined Tesla, his first major job in the industry doing applied engineering with validation, shipping real products and all that mess. During the Model 3 ramp where everyone had to chip in he was down at the factory line, doing whatever he could like everyone else. He has done this for 5 years. Given his rockstar fame and the demand for top DL talent he was probably paid millions of dollars, plus the same in shares that has 10x since he joined. He is very likely set for life... Also he has a CV that is really impressive and can pretty much get any job he wants if he needs more money.

On his free time, he codes the GPT clone minGPT and other small hobby project to understand the latest devevelopment in AI and make it easier for amateurs to access the latest research. He even coded a bitcoin client from scratch just to understand crypto.

A from-scratch tour of Bitcoin in Python

Musings of a Computer Scientist.

karpathy.github.io

Like I said, he is 35years old and, as far as I know, single. When he takes his first holiday, he goes traveling around the world, meeting fellow AI researchers around the world. He also is a mostly vegan person, “90%+ vegan” and he thinks being vegan is one of the best way to reduce CO2.

So imagine a person like that. Likes to travel the world, has worked pretty damned hard and got to the top of his hiearchy, set for life with both cash and stocks. An academic at heart. Maybe he just wants to be free, travel when he wants, do the research when he wants, keep up with latest development in NLP, spend some time on health and do some dating. Heck even explore spirituality, psychadelics and all that stuff that people who travel around the world, are vegan, into crypto etc like to do.

Should he stay longer? He probably has pushed it ”just another year” a few times already. 5 years is a good number to call it quits. Until FSD is released? It has been next year for a few years now. Can he contribute more? He pretty much defined software 2.0 and made the only known large scale implementation of it. The system is set up, it works well. Now it’s time for the applied industry people to run the development from here.

So no, I don’t think we can infer that FSD is near or far away based on him leaving. He probably just wants to have a break, do what he loves and can afford to do and have some more time for the good stuff in life.

 

[Nocturnal]

Think I might crush all my screens and take a break from the world. Nothing but nonsense everywhere. Not even lies anymore, just incoherence. Our ability to communicate is under assault and degenerating across the board.

At least Twitter is down today. That will help slow the spread of stupidity, at least a little bit. Elon should have kept the receipt.

 

[Drezil]

Tesla has top notch AI software engineers and I'm sure they will be able to attract more.

So the loss of Karpathy doesn't mean a whole lot at this point. FSD will continue to make slow, steady progress.

It's the "slow" part that bothers us investors. And the bottleneck for speed is just raw compute power. So I think the development of Dojo is far more important than the development of the kinds of ideas that Karpathy brings to the table.

Depends. Raw compute can beat any algorithm. But the best optimization are in algorithms, not hardware.
And the bottleneck is correctly labeled data for the march of the 9s.

He brought some really cool ideas to the table. Like the 2d-memory-map the network can always only see partially, but move within over time.

The rest was "just" using transformers for recognition, introduction of a very cheap First-Pass so the transformer can be bigger, the reformulation of the problem into vector space (all the nets operate in their eigenspaces, getting the annotations into that was the hard part)..
In hindsight all very "simple" and logical. But hindsight is always..

90% of ML problems are "just" bad/wrong processed data..

On his blog he has a great article on how to approach these things

95% of all big-data ml-tasks can be solved 99% accurate with simple things like K-nearest-neighbor. And many "clever" solutions fail that test no matter the compute you throw at it..
Bur you need them if you want a general solution going above a "what did I do in a similar situation".. because what if it is the first time?!

tl;dr: kaparthy is not needed. Raw compute is not enough. Most important are excellent annotated data.

 

[henchman24]

PPI was also hot… we’re likely to find out if the market really does think that we’ve bottomed. Core was still ticking down slightly.

 

[kbM3]

I

I have FSD (Beta) and while it can do some pretty amazing things .... I'd be really surprised if it was "solved" by the end of the year.

Remember, time frames are not Elon's strongest quality. I'm sad to hear when top level talent leaves a company .... and you HAVE to ask yourself this: IF it was really about to be solved by the end of the year ... why wouldn't Karpathy hang on a few more months to "spike the ball" ....???

Elon didn’t claim FSD would be solved by the end of the year. He claimed it would be safer than human. Although I agree with you that it is doubtful the timeline will be met, it could theoretically be safer than human, and still require nuisance interventions every few miles, and thus be far from ”solved”.

 

[ZsoZso]

Business as usual for her though:

View attachment 828237

So, she states that 70% of accidents of a cars driving under ADAS systems are Teslas, but forgot to mention that 99% of miles driven under ADAS are Teslas -- context matters
All the other ADAS in various cars are barely ever used, because they suck. People try them once and realize the car can't even stay in the lane through a mild curve and never use it again (see Munro's video about the Ford system).

 

[Nocturnal]

So, she states that 70% of accidents of a cars driving under ADAS systems are Teslas, but forgot to mention that 99% of miles driven under ADAS are Teslas -- context matters
All the other ADAS in various cars are barely ever used, because they suck. People try them once and realize the car can't even stay in the lane through a mild curve and never use it again (see Munro's video about the Ford system).

A few months back I tried the Toyota lane keeping in a rental. It was all ping pong and phantom braking. I gave up pretty quickly.

 

[Singuy]

Tslq starts trading today at almost no volume. Guess this short etf has no interest lol.

 

[henchman24]

Tslq starts trading today at almost no volume. Guess this short etf has no interest lol.

I'm so excited to short this short etf.

 

[TheTalkingMule]

Oil is plummetting quite nicely. Headed below $90/barrel.

 

[elasalle]

OPM(Other Peoples Money) trade . Added 2 Jun 24 1K calls, based on proceeds from Oct 22 1100 CC's I sold yesterday
CC's were against recently added Jun 24 leaps in 680's.

 

[henchman24]

And we are now at the yesterday's support....

 

[Usain]

Depends. Raw compute can beat any algorithm. But the best optimization are in algorithms, not hardware.
And the bottleneck is correctly labeled data for the march of the 9s.

He brought some really cool ideas to the table. Like the 2d-memory-map the network can always only see partially, but move within over time.

The rest was "just" using transformers for recognition, introduction of a very cheap First-Pass so the transformer can be bigger, the reformulation of the problem into vector space (all the nets operate in their eigenspaces, getting the annotations into that was the hard part)..
In hindsight all very "simple" and logical. But hindsight is always..

90% of ML problems are "just" bad/wrong processed data..

On his blog he has a great article on how to approach these things

95% of all big-data ml-tasks can be solved 99% accurate with simple things like K-nearest-neighbor. And many "clever" solutions fail that test no matter the compute you throw at it..
Bur you need them if you want a general solution going above a "what did I do in a similar situation".. because what if it is the first time?!

tl;dr: kaparthy is not needed. Raw compute is not enough. Most important are excellent annotated data.

I think I wasn't clear. I'm well aware of the power of algorithmic optimization and I agree with you as far as that goes.

But from everything I've been hearing from Tesla over the last year, compute power is the bottleneck right now. I haven't heard Tesla saying they need more clever ideas. They say the need to iterate faster. And that requires raw computation.

I'm hoping that the recent layoff of human labelers is a sign that the auto-labeling effort is going well. From my understanding, auto-labeling would be a compute-intensive task? Sounds like you would know better than I.

To the extent that they need breakthrough ideas from Karpathy, Tesla might be more likely to get what they need with him working in an academic environment.

 

[82bert]

In case this wasn’t already linked in regards to Karpathy’s departure.

https://twitter.com/x/status/1547341166512771072

 

[UncaNed]

The SP had risen more than 50% in 3 weeks since the split announcement (and 5x since the March low during the COVID dip), so I doubt anyone long TSLA was in danger of a margin call.

I remember it well.

EDIT- Now I see that Tim S referenced the 2020 stock split, not the upcoming 3-to-1, which hasn't been announced yet...

Thanks for setting me straight, @ThisStockGood & @Cosmacelf .
___________________
What do you mean "The SP had risen more than 50%..."??? Wasn't it around $700 on June 10 announcement date, and it's right around $700 now - a 50% rise would need to be at $1,050, no...??

I never quite seem to understand what's going on...