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

[engle]

The man known as "the Leonardo da Vinci of robotics", Professor Dennis Hong of UCLA, posted this on his Twitter feed an hour before AI Day was scheduled to begin:

Dennis Hong @DennisHongRobot 11h

Sometimes your place is on the stage, sometimes your place is behind the scenes. Wherever you are, whatever role you play, we all work together as a team to bring robots closer to our lives, and to bring tomorrow closer today.
@RoMeLaUCLA
Important long term things are often not visible at the beginning but will shine at the end. People with vision are the people who can see through the distractions, and discover their potential and possibilities.

Wise words!

Prof. Hong and his RoMeLa teams' academic research papers have been cited 2,247 times by other authors. When I get my Tesla Bot, I will definitely buy or subscribe to the Tesla Cooking App: (since I only know how to use a microwave )

Task Planning with Mixed-Integer Programming for Multiple Cooking Task Using dual-arm Robot

Here's a huge revenue generator (2020 RoMeLa paper) once Tesla Bot can do it bi-pedal. Note that most packages are less than 45 lbs. which is the "carry capacity" in the preliminary spec. Arm extend lift is 10 lbs.

Alphred: A multi-modal operations quadruped robot for package delivery applications

The rest of this post compares AI Day to my August 4 2021 predictions with my take on some of what was announced.

Tesla AI Day August 19th: Updated and New Predictions based upon Tesla's hint published by Rob Maurer

Tesla Begins Sending Out AI Day Invitations

My 1st prediction that the FSD 4.0 chip architecture and board will be announced was correct.

Prediction #2: it's being installed in all new S/3/X on 8/19 to avoid the Osbourne effect.

Nope. Elon announced will first be installed sometime in 2022 in the Cybertruck -- no Osbourning of Models 3, Y, S, or future X.
Russ' take: FSD v9 is not ready for wide release yet so no reason to put more expensive (not ramped) FSD 4.0 hardware in current production.

Tesla is working on robotics with the "Leonardo da Vinci of Robots" [1]. It should be obvious what the "secret project" is when you check out UCLA Samueli School of Engineering [*] Professor Dennis Hong's lab webpage:

RoMeLa | The Robotics & Mechanisms Laboratory at UCLA

The lab studies: "humanoid robots and novel mobile robot locomotion strategies", including autonomous robots.

Tesla Bot!!

Prediction #3: One of the lab's robots is running Tesla's code running on FSD 4.0 to navigate the real world!! It is funny that one of the lab's robots is called "Darwin" after the scientist of course - not the award winners.

I was overly optimistic - hey, I'm a fanboy since August 2011! Elon claims a prototype will be shown in 2022. Whenever that actually happens it should move $TSLA.

Prediction #4: Attendees on 8/19 will witness at least one of Prof. Hong's UCLA teams' RoMeLa robots navigate the real world running Tesla code on FSD 4.0.

Instead, we saw a dancer in a Tesla Bot costume! I predict Tesla will sell a limited edition of these bodysuit costumes. I will buy two: one to wear at Costco and another one to collect along with my Tesla Short Shorts that don't fit me!

If my predictions are accurate, this will be huge!!

Check!

For example, Tesla Robots will be able to do chores around the house to help seniors live independently with dignity in their homes longer without having to hire outside help. Just like our Tesla vehicles, firmware downloads will enable new features over time. A brand new Tesla product category!!

Check! Agree with others Tesla Bots will be used on the Moon and later Mars.

This is why Elon said during the 2021 Q2 conference call:

"long term, people will think of Tesla as much as an AI robotics company as we are a car company, or an energy company. I think we are developing one of the strongest hardware and software AI teams in the world... So a long story but I think, yeah, probably others will want to use it too and we will make it available."

For insight into what Tesla is doing with the DOJO server technology, one only has to look at what Elon tweeted on 2020-09-20:

View attachment 691907

See what Google did with their generations of TPUs (Tensor Processing Units):

View attachment 691904
Google has announced their TPUv4 but it isn't generally available to Google Cloud customers yet:

Google claims its new TPUs are 2.7 times faster than the previous generation
"This year’s MLPerf results suggest Google’s fourth-generation TPUs are nothing to scoff at. On an image classification task that involved training an algorithm (ResNet-50 v1.5) to at least 75.90% accuracy with the ImageNet data set, 256 fourth-gen TPUs finished in 1.82 minutes. That’s nearly as fast as 768 Nvidia A100 graphics cards combined with 192 AMD Epyc 7742 CPU cores (1.06 minutes) and 512 of Huawei’s AI-optimized Ascend910 chips paired with 128 Intel Xeon Platinum 8168 cores (1.56 minutes). Third-gen TPUs had the fourth-gen beat at 0.48 minutes of training, but perhaps only because 4,096 third-gen TPUs were used in tandem." [VentureBeat]

Prediction #5: Tesla did in fact develop the current fastest DNN (Deep Neural Network model) training supercomputer in the world as Elon predicted a year ago. A product called something like "Tesla AI Cloud" will be announced on August 19th or some time thereafter.

If true, this is also huge for investors!! For example, $AMZN makes most of its profits from AWS [1].

I believe $TSLA will be the first company to solve real-world real-time visual machine navigation (auto, robot, whatever) using their DOJO supercomputers to train their DNN (Deep Neural Network models). The photo shows 25 chiplets per assembly. Those will have the fastest interconnection bandwidth. I expect these supercomputers contain 1,024+ of these assemblies interconnected by some kind of super-fast bus topology. This is DNN training silicon and is distinct from FSD 4.0 which is used for inference only to infer what the video feeds are "seeing" in the real world.

This prediction was quite accurate. Tesla AI Cloud might be announced in the future to generate revenues. Note that Google claims 1 exaFLOPs (one billion gigaFLOPs or 10^18 16 or 8-bit floating-point operations per second) for their TPU v4 ML training supercomputer. Tesla claims 1.1 exaFLOPs.

Tesla Dojo supercomputers can be used for general-purpose machine learning (ML) training. ML training basically involves billions+ of matrix multiplications [2] of vectors and matrices with forward and backward propagation to calculate what are called "parameters" (probabilities) inside a multiple-layer model. The word "deep" in deep learning comes from the fact that often these models have over a hundred or more layers.

Confirmed.

Sophisticated Wall $treet investors have already figured this stuff out. That is what started moving $TSLA higher and broke us back up through $700.

This was pure speculation on my part. Most Wall $treet folks probably have no clue about AI/ML/robotics. It will be interesting to read what some of the so-called "analysts" write about Tesla AI Day. Some may write that Tesla did not have an answer to the NHTSA safety probe.

[1] Dr. Dennis Hong on Achieving the Impossible With Robotic Inventions
[2] How Amazon Makes Money
[3] A Complete Beginners Guide to Matrix Multiplication for Data Science with Python Numpy

[*] My son is an MSCS student at UCLA's Samueli School of Engineering but I don't think he's taken any classes with Prof. Dennis Hong:

View attachment 691908

 

[kanweg]

C’mon Tesla: “humanoid” robots? Make ‘em octopus-like. MUCH more versatile, particularly for fabrication purposes.

Exactly. I thought give it four hands. You can walk on hands. But a robot with four hands can do stuff that no man can. Oh I wish I had four hands if it were not for the fact that long sleeved shirts would be more expensive.

As to unionization it is 3D chess again. 51% of the Tesla workers voted for 24/7 workdays with no salary or bonuses but with sufficient charge breaks.

The realdoll.com people (warning: don’t visit if you’re culturally programmed to reject the idea that people have bodies) will probably be scratching their head about the future.

 

[JohnnyEnglish]

Pay more than the others.
They only need 500K chips. That’s nothing compared to what Intel, AMD and NVIDIA need.

They may well already have them. In the Q1 ER Zach referred to expenditure on custom chips.

 

[SN_8]

Of course it does. Here's a trival example: one Marvin posted at every Supercharger location (starting with the ones used by 24-hr Robotaxis).

Need another example? Robots go clean up Fukushima, BEFORE the cooling water ponds crack and dump their witch's brew of long-lived radionuclides into the Pacific.

Don't let the blinders of your personal prejudices block this clear view of an expansive future. Any tech can be used for good or evil. You wanna stop using fork and knife?

Very good examples. Elon didn't list specific jobs that TeslaBot will disrupt not because he was unprepared but rather because it would create even more unnecessary opposition at this early stage. People are very attached to their crappy jobs (see: coal miners) and he doesn't need to give the UAW or whoever extra anti-tesla talking points.

But I don't think there's a huge lobby for nuclear meltdown martyrs. That would have been a very good example for him to share.

 

[Todd Burch]

So I am assuming the entire Karpathy's talk of the architecture is v10? Seriously doubt there's going to be a significant change for V10 between now and when V10 comes out.

Elon already said there’s a piece of the architecture that will be significantly reworked for v10.

 

[Bladerskb]

D1 (400w TDP, 645mm^2, 7nm)
BF16: 362 Tflops
FP32: 22.6 Tflops
On Chip Bandwidth 10TBps(or 1250 Gb/s)
Off Chip bandwidth: 4TBps(or 500GB/s, 25 D1 per Tile)
Off Tile bandwidth is 36TB/s reported (I think 9TB/s is more like it for tile to tile communication), 3000 D1 chips connected together

AMD Radeon MI100 (300w TDP, 750mm^2, 7nm)
BF16: 92.3 Tflops
FP32: 23.1 Tflops
On Chip bandwidth : 1228.8 GB/s
Peak Infinity Fabric™ Link Bandwidth 92 GB/s (offchip)

Nvidia A100 (400w TDP, 826mm^2, 7nm)
BF16: 312 Tflops
FP32: 19.5 Tflop
On chip bandwidth: 2,039 Gb/s
Off Chip bandwidth: 600GB/s (up to 12GPUs)
Off Chip bandwidth PCI4: 64Gbs

My opinion, the actual D1 chip is pretty good, but not mind blowing given the size/power usage. It's inline with Nvidia's best. But dat scalability holy S balls with them interconnect bandwidth off tile. Mind blowing...

You conveniently left out TPU v4 which destroys all of them.

 

[AudubonB]

Sure, but what will the TMC moderators do all day then?

Grammar is easy (current mix of forum participants notwithstanding). But the day a bot can out-quip, out-pun, out-reference AND out-bringing-experience-to-bear me….that’s when I’ll find out if it can out-sail me, too.

 

[Singuy]

You conveniently left out TPU v4 which destroys all of them.

??? Show Tesla a place to purchase TPU v4 and then we will include it.

 

[Singuy]

Elon already said there’s a piece of the architecture that will be significantly reworked for v10.

If that's the case then Karpathy would have said something at the end. But instead he went into details about reducing overhead cost to decrease latency as ongoing optimization work that needs to be done, not anything about a major architectural change.

 

[Baumisch]

You conveniently left out TPU v4 which destroys all of them.

Any source for that claim?

But after reading/googling a bit through the net - this is a crowded space with lot's of players, NVIDIA and GOOG seem to have the lead, so it's not like the impression on AI Day that I got - which felt like D1 was special - it's special for a "car company" but not too special in the AI/ML space - correct?

 

[Bladerskb]

??? Show Tesla a place to purchase TPU v4 and then we will include it.

Show a place to purchase Dojo and then we will include it.
Heck show Dojo running in an HPC and then we will include it.
Heck show Dojo running on anything other than a test bench in a single node with rudimentary neural network and then we will include it.

There is only one exapod in existence, the other is photoshopped. Hey I can make photoshopped images too and throw specs on there.

 

[jerry33]

Someone looked at the R&D expense difference. If Elon bears down, he could design the electromechanical part of that robot in 2 weeks, himself. The rest is just car parts bolted on.

I think the negativity is from being lied to about how long, and how much money it takes to engineer something. You could be applying NASA costs to SpaceX results.

The problem at NASA is that every Congressman and Senator needs their bit of pork. So the engineers have to redesign every item Congressmen plus Senators times.

 

[nativewolf]

Robots for firefighting? nuclear cleanup? Mars colonization?

A day after Teslabots start rolling off the assembly their will be a startup getting a billion dollars in funding for sexbot conversions.

That is what humans do. The internet will be the greatest tool to bring education to every human on Earth? It delivered porn to Afghanistan, Madagascar and every corner of the globe.

Robots picking lettuce and weeding raspberry fields is more like it. No more immigrant farm labor which will be good and bad. The second gen is what has propelled USA forward.

 

[Singuy]

Show a place to purchase Dojo and then we will include it.
Heck show Dojo running in an HPC and then we will include it.
Heck show Dojo running on anything other than a test bench in a single node with rudimentary neural network and then we will include it.

Sorry are you in the wrong forum? This is Tesla investor forum, where products and hardware TESLA can get their hands on matter. Who cares if no one else can't buy dojo. It does matter if Tesla can't buy TPU v4. Get it? Go troll somewhere else.

 

[hacer]

Standing in the wind, rain and/or snow 24/7/365 waiting to plug in the next robotaxi.

My point was that during the Q&A they were twice asked for what kinds of things might the robot do. Both times the only answer was to repeat
"boring, repetitive, and dangerous" tasks. Of course it's easy to think of lots of such tasks just as many here have done. The thing is that they didn't answer because they didn't want to answer. Announcing even one such task sets a high bar they don't want to set, and for good reason - they haven't started this yet, not even the minimal thinking about the problems so they're not ready to answer. Especially with an absurd 1 year timeline.

Navigating through the world is one thing, interacting with it is another whole ball of wax. There are lots of researchers all over the world working on robotics that "do things" and very little progress has been made. If anyone can solve it, Elon (and the team he will build) can but it will still take a decade or more before it's even slightly general.

Your example is super narrow and might be a good starting point because it is so limited (yet still very difficult), but it's not inspiring so there is no way that would be an answer they could have given last night.

 

[nativewolf]

Anything that involves vision and learning: all doctors and technicians evaluating X Rays/ MRI/ EKG's, ... picking fruits and vegetables, wine making (some high end winemakers already use some form of automated selection of grapes, after they are picked ind cleaned), removing or controlling weeds - so replacing herbicides, ...

I agree completely, obviously you know a bit about wine but the entire agricultural community is starved of labor. Dairy, veggies, etc. The only one this would not impact much would be grains which is already fairly mechanized (gps controlled tractors, image detection, etc). However, no second gen of immigrant labor will hit USA very hard. What is being proposed has long and deep implications for humanity.

 

[jerry33]

Does make you wonder if dining rooms will stage a comeback.

But where would I keep my computers?

 

[AudubonB]

Making robots aesthetically palatable to humans is arguably the most important task.

No one wants machines that look like the ones from The Matrix.

I got a lot of responses to my octopus bot. I was, however, speaking seriously, but INSIDE the box. That is, the gigafactory box. An octopus-like assembly bot, appropriately created, is what the factory of the future needs, not a bipedal quadralimbic one. OUTSIDE the box - yes. Humanoid for human-oriented tasks and interactions.

 

[Singuy]

Any source for that claim?

View attachment 698974

But after reading/googling a bit through the net - this is a crowded space with lot's of players, NVIDIA and GOOG seem to have the lead, so it's not like the impression on AI Day that I got - which felt like D1 was special - it's special for a "car company" but not too special in the AI/ML space - correct?

The D1 is special not because it's the fastest, but for Tesla has brought the cost down by 75% vs buying from Nvidia. Not only that they can tailor everything to Dojo and build the software around it.

Elon said "we have succeeded if we turn off our gpu cluster"..meaning decoupling from Nvidia.

 

[nativewolf]

My point was that during the Q&A they were twice asked for what kinds of things might the robot do. Both times the only answer was to repeat
"boring, repetitive, and dangerous" tasks. Of course it's easy to think of lots of such tasks just as many here have done. The thing is that they didn't answer because they didn't want to answer. Announcing even one such task sets a high bar they don't want to set, and for good reason - they haven't started this yet, not even the minimal thinking about the problems so they're not ready to answer. Especially with an absurd 1 year timeline.

Navigating through the world is one thing, interacting with it is another whole ball of wax. There are lots of researchers all over the world working on robotics that "do things" and very little progress has been made. If anyone can solve it, Elon (and the team he will build) can but it will still take a decade or more before it's even slightly general.

Your example is super narrow and might be a good starting point because it is so limited (yet still very difficult), but it's not inspiring so there is no way that would be an answer they could have given last night.

The answer was given because what they will displace is HUMAN jobs. Jobs that people are willing to do and often the alternatives for them are terrible paying jobs. Many many people are unqualified or incapable of doing anything other than boring repetitive jobs. Picking peaches and apples and pruning blueberry stems is still a job and if you are good at it that is $25-35/hour. Which is not bad in rural USA. What they are proposing will gut millions of jobs from the workforce.

That's life though and the answer was obviously something given considerable thought- still I think they botched it. I'd have focused on the jobs for which we don't have adequate labor. Nursing home assistance where they need someone to cut fingernails or polish fingernails or to just make rounds and wave. Someone to weed organically instead of having to use herbicides to control weeds. Someone to take cows out to the next pasture instead of hoping they find their way there. Basically the answer was a reflection of the past and present. I would suggest that they look forward and to the future. As someone working in rural America in the most dangerous job in the USA I am a bit disappointed at the answer.