Anyone else thought that this line was in reference to Jim Chanos’ radio silence?
Also the peanut gallery shorts on Twitter are now coping by talking about Wework (as if they can short it
)
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
Anyone else thought that this line was in reference to Jim Chanos’ radio silence?
Also the peanut gallery shorts on Twitter are now coping by talking about Wework (as if they can short it
)Already read your post about going back and reviewing your understanding, so kudos to you.
But I'd still like to point out some serious cognitive dissonance in your post. You were trying the justify the performance of Chano's hedge fund as being structural in nature, yet its performance is clearly dependent on his shorts not losing too much money. You mentioned in an earlier post about how it neared break-even.
Think about it, if his short fund was constantly losing 10% annually, his long fund would need to appreciate at least 3.5% just to break even overall. Not having such an anchor against a funds performance sure would look good. So if he could directly affect the performance of his short side, that would juice his fund's overall performance. So his morality directly plays into the mechanics of his fund. It's not at all agnostic.
smorgasbord
Active Member
Some truth? Unless you have numbers to suggest otherwise, what I know today has me standing by almost everything I said in that "old" post. What we all have to realize is that Tesla didn't stand still. The price cuts, variant introductions, option configurations, new markets in additional countries, etc. have all played a role in the constantly shifting demand/sales balance.
I don't think anyone outside of Tesla knows how many LR Model 3s have been sold each quarter in the past 7 quarters. That number fell off a cliff in Q1 of 2019 and never recovered to its 2018 levels. Tesla made it up with SR/SR+ variants (there was a short-lived MR in there, too, remember?). That hurt gross margins, the bleeding of which finally turned a corner last quarter (hopefully for good). Not to mention what has happened with Model S & X sales.
It's great that Tesla survived by pulling these levers, but it would have been better if they didn't have to have been pulled, and pulled so quickly.
EVNow
Well-Known Member
Yes - first time I've heard Dojo described.
So, probably some new HW to make training faster ? Currently their training needs 70k GPU hours.
EVNow
Well-Known Member
There are such prime targets for shorting - not sure why anyone bothers with an unpredictable stock like Tesla. Whole class of companies ...Also the peanut gallery shorts on Twitter are now coping by talking about Wework (as if they can short it
Zhelko Dimic
Careful bull
I thought it was very clear and useful post. ThanksI think we're in completely in agreement about everything, actually. I might just be bad at... words.
Another poster did say about my post that it was "Insanely obfuscatory and absolutely ridiculous".
The good videos of that scene generally try to stick reasonably closely to Hitler's original script but change it up in hilarious ways. If you just have him saying pretty much whatever then it just isn't that funny. Any video which actually manages to make him mention Stalin in context with whatever the video's joke script is about when he clearly shouts Stalin always gets an A+ in my book, for example.
EVNow
Well-Known Member
I posted this in the autonomy forum. Looks like the HW3 upgrades are routinely happening now when you schedule some other service appointment. The second paragraph is a quote from X sub-forum.
Looks like it takes about 1.5 hours to upgrade to HW3 and install new software. If they figure out a way to make the software update quicker (may be using USB, instead of download) - they can complete the HW3 upgrade in 30 minutes.
"I setup a mobile appointment to fix a leaky rear taillight on my X recently. The Mobile tech installed hw 3 as a "work with". It took him 20 minutes to install the hw3 and an hour to download and install the right software version. "
"I setup a mobile appointment to fix a leaky rear taillight on my X recently. The Mobile tech installed hw 3 as a "work with". It took him 20 minutes to install the hw3 and an hour to download and install the right software version. "
anthonyj
Stonks
Very bullish af. Means they want to recognize that revenue this quarter and release features
EVNow
Well-Known Member
I would not go that far - just that they are fine with the revision of HW3 they have and they want to reduce upgrade cost as much as possible.
Ultimately they will have to do full upgrade only service appointments - that's when the big bucks hit the P&L. As Musk said he is hopeful City NOA will be in early access this quarter. All events point to a Q1 release & recognition of revenue - which would also help the seasonally low quarter.
They need Starlink to speed up the data download!
Thanks! Super interested in these videos, watched all of AKs videos! =)
Guess the new information was the EKF-SLAM style occupancy grid which I will try to analyze more, the sonar and camera road edge visualization was cute, but felt a bit dated compared with the rest. The neural network deconvoluted map was really cool, felt very modern and interesting! But my main takeaway is that they use the multitask CNN to do a lot of object detection etc and use the high level features of this neural network as input to train and to run a RNN. Which feels a bit similar to OpenAI-five:
Note the LTSM 1024 units to the left, which would be the equivalent of this RNN and the input to this network would be high level features of the AKNET_V9 style multitask network.
This was new information. Pretty cool, but not super happy that Tesla are telling the rest of the industry the correct answer. Imo it seems like a really good solution.
Also liked the go on holiday and come back to improved performance quote. Imo that is something awesome with Tesla’s software 2.0 solution, once it is up and running it can only be improved and there are so many ways to work to improve it and the effects will ripple out to all parts of their software 2.0 stack.
Project Dojo seems to be a custom system for training Tesla’s neural networks. I wonder how much own hardware they have developed for this, I assume that Elon Musk gave Pete Bannon two task:
1. Develop HW3
2. Develop hardware for Project Dojo
We know how well he did 1 from the Autonomy day. I would guess he did 2 equally well.
I wonder how much these cost, would not be surprised if Elon gave a green light to split the budget 50/50 for these and that Pete Bannon has built some pretty massive amount of very specific hardware to train Tesla’s neural network. Don’t think that these will compete with Alphabet’s TPU 3.0, but maybe they will perform better for Tesla’s specific needs. If so Tesla’s moat might just have gotten wider. Maybe they could compete, if so maybe Amazon/Microsoft/Google could be customers/competition. Would be pretty crazy if Tesla decides to enter the cloud compute market at some point. Hyperchange should make a video about this! =)
To be followed!
Last edited:
Fact Checking
Well-Known Member
I think your post back in March captured what happened in Q1 pretty well - much better than most of the commenters, including me. Q1 was a perfect storm of various factors and the stock price was hurt badly. So kudos for that call!
But it baffles me that you are still insisting on that post and its approach in today's context. The SR+ likely has margins beyond 15%, Tesla achieved this via the various price increases, bundling decisions and scaling up volume which reduces cost of goods.
Margins will continue to improve as Tesla scales up internationally and increases the utility of 100% margin options, like FSD.
Yet you still insist on LR take rate being the primary metric of Tesla demand, ignoring SR+ sales and improving margins. Why?
Also, how are you feeling about bits like this:
Wasn't the "there's no SR battery pack, it's just a software limited MR" a pure, 100% TSLAQ conspiracy theory?
The SR+ battery pack made by the new Grohmann Machine obviously exists and was a great factor in reducing costs and improving margins.
Last edited:
Artful Dodger
"Neko no me"
No, that statement is a clear reference to the ever-growing 'TSLAQ Block List' on Twitter.View attachment 475159
Anyone else thought that this line was in reference to Jim Chanos’ radio silence?
Also the peanut gallery shorts on Twitter are now coping by talking about Wework (as if they can short it
Elon has poked at a raw wound here. Expect salt next. Then the coupe de gras...
Last edited:
Fact Checking
Well-Known Member
Yes, Karpathy specifically described Project Dojo as Tesla's own neural network training chip: "doing the same order of magnitude speed improvement on the training side, as we did on the inference side with HW3, at lower costs".
Internally they jokingly refer to Project Dojo as "Vacation Mode": it will allow them self-improving neural networks where training is fully automated and iterating so fast and so reliably that the AI team can go on vacation. (While the labeling team still labels images of course.)
Other interesting tidbits in Karpathy's presentation held at around October 10 (a month ago):
- The HW3 board is cheaper to make than the HW2 board,
- Smart Summon had ~700,000 uses two weeks after release.
- He described the size of the HW2+ AutoPilot fleet as "almost 750,000 cars".
- Nice top down "virtual LIDAR" data representation + videos in the presentation, "stitched together" real-time from 8 cameras by the neural networks themselves in the car, not by C++ code.
- Labeling includes traffic signs and road signs already.
Fact Checking
Well-Known Member
I think Karpathy already telegraphed that they are converting their C++ driving code into Software 2.0 code a year ago - where RNN's (stateful CNN's) are the most generic toolkit.
The important bit where hardware accelerated CNN's are required is processing the insanely high rate ~2.6 GByte/sec input data stream (800 frames per second), mainly generated by 8 cameras.
Once that mindbogglingly intense raster data stream has been converted into a rich stream of much smaller high level object data, RNNs running on general purpose ARM cores is the obvious next step - Karpathy didn't really give away anything there.
Tesla could probably even open source all their NNs, and their competitors could still not catch them: Tesla has a fleet of soon to be 1,000,000 cars that are sending them training feedback all around the clock. This is the real NN training computer Tesla has built - which hardware of 1,000,000 cars has to be owned to run Tesla's architecture...
Nobody can match Tesla's neural network iterative training economies of scale at this point I believe. Game, set, match?
Last edited:
I also loved hearing about Operation Vacation. The Autopilot team is trying to automate away their own job so Autopilot can improve itself just with new data.
Buckminster
Well-Known Member
Similar threads
- Locked
- Locked
