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

[JRP3]

Unfortunately diving into that is beyond what I can confidently say, but on a physics stand point there is a few reasons

The point I'm trying to make is related to the physics of ion diffusion through the SEI layer and into the graphite structure as shown in your linked video around the 10:40 mark. This is why charge rate is lower than discharge rate and is the limiting factor not the tab structure.

 

[heltok]

Imo Tesla will go deep into LLMs. I was thinking about the latest GPT4 paper. At page 9 there is this image:

Then it hit me that this is not too far from what Optimus will be doing. Input the images -> compress them into some vector and use as input for the model.

Replace the user prompt with
User: Grab me that can of coke, open it and pour it into a glass

Replace the GPT4 output with:
At XYZ1 there is a can of coke that the user is pointing at
XYZ2 there is the shelf with glasses
Execute list of tasks:
1. Move closer to shelf
2. Select glass
3. Grab glass
...

That's it.

And also in order to interact with the user they need a good speech recognition. OpenAI has released Whisper which runs pretty fast:

Introducing Whisper

We’ve trained and are open-sourcing a neural net called Whisper that approaches human level robustness and accuracy on English speech recognition.

openai.com

You don't need crazy offline compute for this, you can run some of these LLMs on modern laptops:

Running LLaMA 65B on a 64GB M1 MacBook Max with llama.cpp

lawrencecchen.com

So what does this mean? I think Tesla has made upgrades to Dojo to better handle LLMs in the future:

https://twitter.com/x/status/1644659294153977858

And they will be needing these to train their models. Both their massive offline models and their destilled online models. And process so many user interactions and quickly iterate on these huge billions/trillions parameter models. And I think HW5 will have some more LLM specific architecture, not just vision batch size of 1.

Imo Tesla needs to get onto this soon or OpenAI will do with them what they did to Google with ChatGPT:

Ilya commented on this a while ago:

Basically it's all about being willing to bet big and get to scale. Before they were not ready and digital was easier, but now they are getting ready. Elon understands this and is crazy enough to try. And as Elon said, if anyone should get to AGI it's probably best if it's Tesla as they are a public company and he doesn't trust the rest:

 

[GSP]

True, so lets look at any chemistry and the reasons that discharge rates are higher than charge rates. Certainly it's beneficial to keep internal heating within an acceptable range but the limitation in charge rates seems to be more tied to the diffusion rates of the anode than internal resistance of the tab structure.

Exactly.

For those not familiar, charging faster than the diffusion rate the anode can support results in Li metal plating on the anode. Li plating results in a permanent loss in energy storage capacity, and if allowed to continue will produce dendrites than can puncture the seperator, resulting in a short circuit and a “thermal event.”

Charge rates will still have this limitation with tabless electrodes , but they still will provide other benefits by reducing heat losses.

GSP

 

[Cosmacelf]

Imo Tesla will go deep into LLMs. I was thinking about the latest GPT4 paper. At page 9 there is this image:
View attachment 926647

Then it hit me that this is not too far from what Optimus will be doing. Input the images -> compress them into some vector and use as input for the model.

Replace the user prompt with
User: Grab me that can of coke, open it and pour it into a glass

Replace the GPT4 output with:
At XYZ1 there is a can of coke that the user is pointing at
XYZ2 there is the shelf with glasses
Execute list of tasks:
1. Move closer to shelf
2. Select glass
3. Grab glass
...

That's it.

And also in order to interact with the user they need a good speech recognition. OpenAI has released Whisper which runs pretty fast:

Introducing Whisper

We’ve trained and are open-sourcing a neural net called Whisper that approaches human level robustness and accuracy on English speech recognition.

openai.com

You don't need crazy offline compute for this, you can run some of these LLMs on modern laptops:

Running LLaMA 65B on a 64GB M1 MacBook Max with llama.cpp

lawrencecchen.com

So what does this mean? I think Tesla has made upgrades to Dojo to better handle LLMs in the future:

https://twitter.com/x/status/1644659294153977858

And they will be needing these to train their models. Both their massive offline models and their destilled online models. And process so many user interactions and quickly iterate on these huge billions/trillions parameter models. And I think HW5 will have some more LLM specific architecture, not just vision batch size of 1.

Imo Tesla needs to get onto this soon or OpenAI will do with them what they did to Google with ChatGPT:

Ilya commented on this a while ago:

Basically it's all about being willing to bet big and get to scale. Before they were not ready and digital was easier, but now they are getting ready. Elon understands this and is crazy enough to try. And as Elon said, if anyone should get to AGI it's probably best if it's Tesla as they are a public company and he doesn't trust the rest:

I agree. As soon as I saw LLMs, I thought “that’s the missing piece for Optimus”.

 

[cliff harris]

I don't think the likes of OpenAI are real optimus competitors. In my experience of silicon valley coder types, they are VERY software focused. The idea that the software may actually run on an actual physical piece of kit in meatspace is just alien to them. The overwhelming majority of software engineers in the valley have never done any manual work of any kind, at any point. They are not the sort of people who ware going to leap into working all-day-long with engineers and hardware types.

I think a much more likely competitor would be an existing robotics company. ABB or KUKA or whoever it is who owns Boston Dynamics this week. In general, people seem to get overexcited by software tricks, and underestimate the difficulty of stuff like getting a robot to walk, or lift something, or throw something.

I wouldn't be surprised to see OpenAI, Micrsoft,Google, to go the 'AI-directed, human-in-the-loop' direction instead, where AI, maybe using Augmented reality, basically uses minimum wage humans as meat-puppets to carry out tasks. Like amazon's mechanical turk, but on steroids.

 

[philw1776]

Megadittos. I agree that silicone (sic) valley et. al. types are hardware averse. Hardware makes one reality cognizant. Only opinion difference is that I say that Boston Dynamics bots have little connection with reality in that they are hand programed (used to be hard wired in the olden days) to execute their routines. Big opportunity for Tesla here as they are accustomed to TRYING to write code to interact directly with physical reality. The challenges are legion. hope my Tesla investment is a winner here.

 

[generalenthu]

I don't think the likes of OpenAI are real optimus competitors. In my experience of silicon valley coder types, they are VERY software focused. The idea that the software may actually run on an actual physical piece of kit in meatspace is just alien to them. The overwhelming majority of software engineers in the valley have never done any manual work of any kind, at any point. They are not the sort of people who ware going to leap into working all-day-long with engineers and hardware types.

I think a much more likely competitor would be an existing robotics company. ABB or KUKA or whoever it is who owns Boston Dynamics this week. In general, people seem to get overexcited by software tricks, and underestimate the difficulty of stuff like getting a robot to walk, or lift something, or throw something.

I wouldn't be surprised to see OpenAI, Micrsoft,Google, to go the 'AI-directed, human-in-the-loop' direction instead, where AI, maybe using Augmented reality, basically uses minimum wage humans as meat-puppets to carry out tasks. Like amazon's mechanical turk, but on steroids.

You're misguided if you think large companies like Amazon, Microsoft together with what OpenAI are not workign on this. You just have to watch the video below.

The design may have a slightly different line of thinking than optimus, but they are surely thinking of this.

 

[Mengy]

Tesla’s growth is still firing on all cylinders (pardon the pun). I count at least four major greenfield projects going on.
1. GigaMexico
2. Tx Lithium refinery
3. GigaNevada Semi and battery factory
4. Shanghai MP factory.

Tells me Tesla expects to generate copious amounts of cash flow this year and next.

What's even more exciting is most of these positive catalysts will be coming online right about when the Fed is lowering interest rates in the US and the market / economy starts improving again, which should equate to a market rallying right when Tesla's financials are bursting at the seams.

The future is so bright I have to wear shades!

 

[oldTAVguy]

This may be bigger than HUGE...

https://twitter.com/x/status/1645064550725701635

Chaos in Dealership land?

 

[oldTAVguy]

This may be bigger than HUGE...

https://twitter.com/x/status/1645064550725701635

Chaos in Dealership land?

The dealerships are the foundation upon which Ford, GM and other OEMs are built. Inventories at the OEMs could have to rise significantly in a world of high interest rates.

 

[Fred42]

This may be bigger than HUGE...

https://twitter.com/x/status/1645064550725701635

Chaos in Dealership land?

I thought the OEMs' financial subsidiaries finance dealer inventory.

 

[InDaClub]

Lots of CyberTruck job postings

Tesla Careers

Careers Search

www.tesla.com

 

[JusRelax]

Jeff Roberts says 50 MY per hour in Austin... if so, the ramp at Austin is significantly accelerating.

https://twitter.com/x/status/1645185574674874368

 

[wipster]

Sounds like anyone with dry powder should spend it quickly...
Lots of positive developments in the last couple of weeks...
Could be a nice bump coming soon.
Or I could just be hungover and my brain is off duty

Too many eggs already mate! They hardboiled your head!

 

[Cosmacelf]

I don't think the likes of OpenAI are real optimus competitors. In my experience of silicon valley coder types, they are VERY software focused. The idea that the software may actually run on an actual physical piece of kit in meatspace is just alien to them. The overwhelming majority of software engineers in the valley have never done any manual work of any kind, at any point. They are not the sort of people who ware going to leap into working all-day-long with engineers and hardware types.

I think a much more likely competitor would be an existing robotics company. ABB or KUKA or whoever it is who owns Boston Dynamics this week. In general, people seem to get overexcited by software tricks, and underestimate the difficulty of stuff like getting a robot to walk, or lift something, or throw something.

I wouldn't be surprised to see OpenAI, Micrsoft,Google, to go the 'AI-directed, human-in-the-loop' direction instead, where AI, maybe using Augmented reality, basically uses minimum wage humans as meat-puppets to carry out tasks. Like amazon's mechanical turk, but on steroids.

There won’t be an Optimus competitor for at least five years. No one else is building a mass manufactured robot, using optimized in-house actuators, and knows batteries and vision systems as well as Tesla.

Note that, in this case, Tesla is attacking a green field opportunity. There are no entrenched competitors. On the other hand, it is not a slam dunk that Optimus will be a successful product. All depends on the price, specs, how good the AI is, and performance capabilities. I refer you to Tesla solar roof for an example of a product that didn’t live up to expectations. I personally think Tesla can pull off Optimus as a great product, but it isn’t a given.

 

[ThisStockGood]

Elon following the PM of India… hint hint…

https://twitter.com/x/status/1645299269195247618

 

[Tes La Ferrari]

Tesla Supercharger Network hits 45,000 milestone worldwide

The announcement of the Supercharger Network's milestone comes less than five months after the installation of the 40,000th Supercharger.

www.teslarati.com

TLDR: Less than 5 months after installing their 40,000th charger, Tesla surpasses the 45,000 mark for installed supercharging stalls.

 

[BioSehnsucht]

I wouldn't be surprised to see OpenAI, Micrsoft,Google, to go the 'AI-directed, human-in-the-loop' direction instead, where AI, maybe using Augmented reality, basically uses minimum wage humans as meat-puppets to carry out tasks. Like amazon's mechanical turk, but on steroids.

Well, I guess the upside of meat puppets is it that while it will still displace humans in the workplace, it will displace them to menial tasks (rather than entirely removing them), so the eventual costs of UBI (to prevent sending millions to live and die on the streets) might be able to be reduced? (/s, but only kinda)

 

[Optimeer]

Jeff Roberts says 50 MY per hour in Austin... if so, the ramp at Austin is significantly accelerating.

https://twitter.com/x/status/1645185574674874368

We will find out about this new weekly record production rate very soon, at the earnings call the latest. To manage expectations, it is still possible that this is a stress test week to detect the current bottlenecks etc., so a bit lower production rates for next weeks are possible.

The increased production rate of the Model Y for the US market explains the recent price reduction for the Model Y RL and P in the US.



It looks also like the 4680 ramp in Austin is progressing well. During the Q4 22 earnings call one of the 4 lines was running, while the other 3 were in the installation phase and the comissioning phase. Now probably at least one additional 4680 Line in Austin is in production.

 

[UkNorthampton]

Anecdotes regarding car finance tightening in USA (eg Capital One - "sophisticated"), delinquencies generally ok but probably getting worse. Smaller/dodgier dealers (especially used cars) may have fewer (more expensive) funding options both for their car inventories and customer purchases.

My thoughts more generally

1. Some incumbents' profits might be getting better as their car sales volumes level or decrease
2. Fast growing Hyundai has low profits
3. Might some of this be that legacy OEMs rely on high profits on spare parts sales on older models? Paradoxically profits improve as car sales shrink. Less overtime/wages/parts costs at a time when labour & material shortages abound but this would be offset by fixed costs/utilisation becoming larger per-unit. Perhaps there is some temporary advantage from slowing production, Hyundai having a profit of around $800 per car (world markets, many cheap cars) but growing sales means less of the higher profit out-of-warranty spare parts per new vehicle.
4. UAW & other unions more militant, strikes expected
5. Tesla volumes increasing
6. New electric better than used ICE? Better being subjective, includes elements of price, maintenance, fun, social - relevant to each buyer.
7. Unionised OEMs in USA may not be able to switch to EVs or even produce enough ICE during a short period, more customers contemplating alternatives, EVs, Teslas - open door for available vehicles, especially Model Y & F150 type vehicles.
8. Unionised OEMs may concentrate on highest profit specs, above many people's budgets.
9. Harder financing for dealers & customers, lots of inventory to finance at higher interest rates.
10. More awareness of EVs & Teslas in particular, Supercharger advantage.
11. I think there will be unexpected beneficial tailwinds for Tesla ahead, not that they need it, but still welcome as it accelerates the mission.
12. Awareness of Tesla advantages may cut through perma-FUD (a little), to stock price.