Agreed. They have a wide range of impressive engineering activity.

I would say 2023. Wall Street is not going to be so smooth with it. But yes, it will be the most valuable company on earth (literally)

I don't think Wall Street will react for a while. Tesla just announced that they are a HUGE company when it comes to tech. The tile alone has more compute power than 460 of the TOP500 supercomputers. They just showed us that they are better chip designers than INTEL, AMD, NVIDIA that they are better at AI than Google, MobillEye, Waymo, that they basically demonstrated exactly what they wanted to. If you want to work on cutting edge AI, cutting edge electronics you need to work for TESLA. Wallstreet will only realise this in 6 months or so when their tech guys write their little report.By that time the Texas and Berlin will be up and running and the new stock price will be attributed to the new factories. Most of Wall street think they are dealing with a dog. After all, it has 4 legs and a tail. Meanwhile what they are seeing a wolf.

@@jacobuserasmus please correct me here but seems like Dojo has more than 2xCompute power (1.1 ExaFlops=1100 PetaFlops) compared to current worlds most powerful super computer Fugaku (442PetaFlops).

@@rakeshbandi2840 Yes DOJO will have considerably more processing power than the worlds fastest computer and it will fit in a small data centre. DOJO's applications are ridiculous. If TESLA shutdown everything else and just sells DOJO they will be living large.

I thought they were just trolling us at the beginning of that part.

Military have been developing it for long time, currently using exoskeletons to compile data.

I agree with you somewhat but this event is for hiring, so I think they just want to attract people who want to be part of the project.

This event is for hiring, you need to throw a concept to interest talents. This is a product announcement event.

I think they have all the pieces. Especially the hardest piece. FSD bits with hyper scale auto labeler. The hardware essentially is miniaturization of existing power electronics and battery tech they already know inside out.

Teslabot!=boston dynamics. They never took off because they could not pull off better batteries and AI to deliver real world robots. So they’re stuck with demo dogs on youtube sponsored now by new Korean overlords that will try to apply that to industrial scenarios. Teslas has the hardest bits that have so far evaded everyone - real world AI bits that generalize and hyper scale. Teslabot will have profound consumer and industrial applications. Just in time as world depopulates. Imagine alone how many seniors can benefit from having a useful robot that can wash dishes, pick up mail, walking/security companion. Finally the end of dumbass Roombas defining what real world robots can do. They’ll pull off something the Japanese have been trying for decades - but bounded by their limited software skill set.

Machine learning and AI Scientist here, let me try to explain why thy are confident they can get a prototype bot out by next year. 1) They have fundamentally solved the twin problem of vision and planning for navigation. All the remains is scale based training to identify and learn to react to edge cases in the world space. Adding more "nines" is now an arbitrary task and it doesn't require an exponential ramp in processing cost to cover vastly more landscape of the action space than any human driver. Expect 99.99999% accurate driving within 2 years. 2) The nature of their solution can be readily shifted to other frames. The cameras are agnostic to the frame they are within. The abiliity to integrate camera views into vector space for plotting position, planning and navigation will work on a bipedal frame with two or four cameras as "eyes" in it's head coupled with pressure sensors and actuators and motors to drive ambulation. 3) Training the frame to walk, run , crouch ...etc. is a matter that can be handled purely in simulation. Unlike the driving problem which involves a significant number of non static interaction scenarios (other moving cars) ambulation of a bipedal frame is mostly focused on the environment....where interaction with dynamic agents is present, navigating to avoid obstacles or crashes is a vastly more trivial problem at low velocity interactions compared to cars. Not only will inferencing be computationally faster it will not be as costly. 4) Distributing the Neural nets through the physical frame will be the key to dramatically collapsing complexity and increasing high fidelity. For example, the hands can be driven by a model trained to manipulate a range of items in simulation. This can be loaded into the FSD and invoked when certain tasks are requested by executive function (ie the goal) for example, likewise heirarchical networks specializing in solving motion tasks ...say to get the hand to a cup on the table can be loaded into the arm, thorax, hips, knees and feet. In fact these problems have already been solved for certain simulated and real world robots (see the company in China that has created a "Spot" like Boston Dynamics but uses neural networks instead of hand crafted statistical learning to dramatically reduce the mechanical cost of the frame (they are aiming to sell it for 5000 compared to Boston Dynamics 70,000 Spot as I recall). All told, given the level of integration and process efficiency Tesla is demonstrating toward the much harder Self driving problem...I don't think it's unrealistic that they can have a prototype by next year of Optimus.

George hotz could still win self driving. Tesla might flounder in the long tail of edge cases for ages while comma, the tortoise in this race, wins because of their superior generalisation.

@@singularity844 we will see . I believe in Andrew and his team

@@oxide9717 yeah well that’s a safe position because they are the front runner!

i like tesla but i also like driving😥

I mean it was a false dichotomy. In other words, a bad question

The neural net is not being trained to kill the passenger OR the pedestrian. It's being trained to 1. Perceive the environment 2. Drive safely and efficiently from 1 location to another. Things move quickly to humans, but this is because our reaction speeds did not evolve to drive a car safely at 65 mph speeds. Things move slowly to a computer who sees and processes frames almost instantly. Tesla's system processes 2,300 frames per second. The human brain can see perhaps 30 or 60. I can prove that this Is a false dichotomy on a code level and on a modular hardware level as well, but what you need to know is that nobody is focusing on who to kill. They're focusing on making the system as good as possible so that nobody has to Die in those situations where a human would have to choose. There's nearly always a way to avoid collision and fsd will be much better at this than humans. Humans kill 1.3M people per year in cars accidentally. I suspect that in 15-20 years when nearly all cars are autonomous and electric that less than 1000 people will die in auto accidents which is a 1000× improvement. I hope this helps.

@@pranjal86able , As a machine learning practitioner I've been asserting that there is no need to teach moral decision making to a FSD because all you need to do is make sure it obeys traffic laws better than any human. When that is the case such a vehicle which it's extended visibility of the roadway can react by stopping ahead of any scenario including intersection violations like drunk drivers of teenagers joy rides. Further, as more cars on the road become FSD capable the accident probability will be crushed down dramatically. It will be in the best interest of humanity for cities to pass laws to actively ban "free driving" on certain roads at certain times... as I called it 6 years ago. Even smarter cities will ban human driving out right indefinitely as they start seeing that as more of the cars become part of the tesla fleet , not only will they drive more safely but they will be able to do it at faster speeds (up to the limits of the law). This will allow road speeds to be raised in order to increase throughput at peak times at a time where the number of transport events will be higher than they are today (much higher as people will be getting into SDV's for everything and there will be a lot more single passenger transports)

@@DavidSaintloth That's the most reasonable, forward-thinking comment I've seen on YT in a long time! Thank you for not being a blind believer of the mainstream media like most people. 🤣

They would correct like every one else ... however because their economy of scale and their cash hoard are beyond the competition they will suffer less if any pain during the dip. In fact, if they succeed in crushing costs due to their high production efficiency they can be seen as the only viable option for those looking to purchase vehicles in a depressed economic sittuation...this in fact happened to a degree during the covid year of 2020. Finally, once the next quarterly report shows their delivered production numbers and progress in Tesla power and batteries would all open eyes as there is likely to be a big ramp in switching to green tech during a down turn. All and all the answer to your question is YES.

I'm happy to wait a bit longer for my pre-order if it means no requirement to go back for a computer upgrade at any point. I don't expect the Aussie Cybertruck to be available until at least 2023, so no big deal waiting a bit longer. More time for my shares to go up.

HW can be upgraded later like they did with model 3. Cybertruck can arrive earlier.

The Amazon Optimus Prime would have been the perfect name.

I want a break dancing robot.

Patience Karthik - Dave's busy st the moment with other priorities...

The devil is in the details, my guess is that it would take years to copy from scratch

Maybe Tesla's fundamental competitive advantage is its capacity to innovate- not the technology itself. Tesla's AI day is really a pitch to the world's foremost AI talent to come on board. By advertising their accomplishments in this 'promotional event' and securing further talent, i'm guessing Tesla see a lot of upside rather than downside. This contrasts greatly to the Buffett-style traditional moat which focuses on locking down IP and achieving unassailable market scale (ie a competitive strategy which seeks to monopolise customer choice as opposed to Tesla seeking to extend customer value.

He said a few years ago Google was the clear AI leader.

Not really but they are getting there

Hi

Ya tesla should just copy a human

one more thing that can break though, it adds so many new things to worry about and you can do the same thing with multiple / wide angle cameras. In the words of Elon "the best mechanism is no mechanism"

And cameras can be static since the pixel density is so high. But could be cheaper indeed if it could move ;)

If a wide enough field and enough resolution, the software essentially zooms in on interesting areas. So, moving cameras add complexity and cost with little value.

multiple cameras stitching 360 degree view is done in software, no mechanical moving parts, no added points of failure. Less complex and cheaper.

Fewer vehicles,fewer accidents with robo taxis.Less energy as we don't need many vehicles.

@@srinimaram And the humanoid robot does what?

@@bmet001 the robot will have your job so you dont have to work unless you want to

@@timothyhodgkiss9051 and that accelerates the world to sustainable energy how?

@@bmet001 use your own brain to figger it out, or do you need to grow brain sells too, if not ask elon for a neuralink implant im sure you would be one of the first to need it lol........................