I drove to Red Lodge and back yesterday on V12 with the menu selection of letting the car decide how fast it goes. I hadn't selected that before and it makes a big positive difference. Coming up to a single lane construction zone with a flagman, it came in a little too hot for me so I braked to a slower speed then put it back in auto. It then slowed nicely to a stop at the flagman where I was first in line. The stop sign flag with slow on the opposite side was on a stand allowing the flagman to not hold on to it. As the opposing line of cars just barely finished coming through, the flagman reached for the flag pole to turn it and even before turning it, the X (very comfortably and not slowly) started to pull around the flagman and go. Just like a human would.
Thank you guys. It is great to hear two experts discussing this complex topic - instead of a clueless "reporters" who just can't understand the concepts. You have a synergy as friends and colleagues that should be a model for technical content.
I think, Optimus will penetrate the nursing home market before the consumer home. Nursing homes can be equipped with complementary equipment and room layout to accessorize Optimus in a very controlled environment. As accessories become available (aftermarket) so the medical corporations have the $$ to invest. Optimus supervisors (RN’s hopefully) could manage so many more patients, yet provide healthier environments for patients. ER rooms also. An RN would more effectively train the bot. Cheers from San Diego.
I assume T-bots will flow to where they produce the highest returns for a company. A nursing home is going to gain a lot by having extremely reliable, train-them-once "employees". The average annual wage for LVNs in California nursing homes is $60,855. It takes between 3 and 3.5 workers to cover a position 24/365. That's over $180k per year plus an additional 30% or so for employer overhead. Perhaps more when calculating in recruitment, training, etc. The value to a nursing home would be based on avoiding roughly a quarter million dollars per year in staff costs by buying or leasing one T-bot. Few households would be willing/able to outbid a nursing home.
10:20 When it runs out of power... I recall a SciFi story - the locals used steam engines for motive power and imported chips for control. A robot serving drinks started to wobble and the owner shouted to it "Stoke you idiot! Stoke!" and it staggered over to the fireplace, and shoveled a few lumps of anthracite into its firebox to restore steam pressure! Possibilities!
Not only is Elon going for the factory market, he is going for his own factories first. Do other manufacturers have anything like that? Any problems Tesla encounters with Optimus in its own factories will be in house problems. Any problems encountered in other factories will NOT be in house problems. There is a world of difference.
@14:41 *Human:* _What do you plan on doing with that knife?_ *Robot:* _I plan to dice the vegetables, slice the potatoes, boil them, cut the bread, sautee the steak, set the table,_ _and if you don't stop pestering me with your asinine questions, I will feed you your own liver with some _*_fava beans,_*_ while I sit back and have some _*_Chianti._*
An order of magnitude more difficult than training a bot on a task is training a bot how to be part of a team -- a team that may include other bots and/or humans -- where there isn't a 'clean' situation. This is similar to FSD driving on a road by itself (100% solved) versus driving in the real world where stuff happens (99+% solved). This is where Optimus should excel.
Robots working in a room or restricted space has the advantage of many cameras watching them. The footage would be realtime processed by a powerful backend AI that does the "digital twin" simulation , thus able to predict with more accuracy each individual bots working. It will relay "overide" commands in realtime when needed to the bots, thus no collision or clashing on the workplace. Those learned model will periodically be incorporated to each bot so the llearned data not from Bot is absorbed.
@@fteoOpty64 That sounds more like the interface planning for KUKA robots. What if there are people in the space? Optimus needs to naviagte the space just like FSD navigates the roads... courteously, carefully, assertively, all in the proper time.
Telsa's FSD goal is level 5. Telsa's Optimus humanoid robot goal is full sentience. This is why Elon wants 25% voting control because his goal is to achieve something that he himself stated has a 10 to 20 percent chance of destroying humanity.
@@richb2229 In the conference call Elon used the word sentient not symbiotic. " I've said before I think Optimus will be 24:49 more valuable than everything else combined um because if you've got a a 24:55 sentient humanoid robot uh that is able to navigate reality and do tasks at at request um 25:05 there is no meaningful limit to the size of the economy.
A G search gives some conflicting info on HW-4 and HW-5. Samsung is the primary producer, but Tesla was messing with TSMC chips. I suspect Samsung is working hard on power usage. I think most Tesla cameras are Samsung. Samsung building a big plant NE of Giga Texas seems important.
Yes V12.2.4 is so much more polite. I did have issues with yellow green cones twice. It would stop in front of the cones and stop making me look like an idiot and I would have to wave my arms to go around me. This happen twice on the yellow green cone where some workers had painted the street. To get out I had to back up and go around the cones. There was no danger at anytime.
Tesla could use their employees to train Optimus. Elon could say, 'Hey, we want to train Optimus in a home environment. Anyone want to take one home?' Tesla has thousands of employees. How many do these other companies have? A couple hundred at most? There an advantage there.
Tesla - "Take a robot home, and train it." Employee - "Ok, no problem." (One week later) Employee - "Hey, something's wrong. My ID badge doesn't work anymore." Tesla - "You've been fired. The robot has your job."
About privacy thing it is good point. Even when selling only hardware there may be hardware backdoors so that those bots can communicate satelites quite easy and you dont notice anything also you canot verify model as easy as software
When the bots are moving 10x faster and more fluidly and they put those bots on the assembly line to make more bots... You'll have Botception. The wait until they start Gigapressing bots. Holy crap
I'm not as worried about Optimus destroying humanity as about humanity destroying itself. In my lengthy lifetime we have come perilously close to doing that on our own. We have lived under a sword of Damocles for so many years we have come to accept it as unlikely, rather than inevitable. Optimus may instead save us from ourselves, or at least remain behind to tell our legend.
Hey John sitting and listening to your video and thought... If Tesla want to get voice to text, then all employees could get a microphone for training and relevant phrases
Just curious, why would the humanoid bot use 1 chip when the car uses 2? What is the reason for using 2 chips in the car, is it for redundancy? Excellent video!! Starting it off with that joke put a smile to my face
Apple was already leveraged in the education and financial markets before it ramped into the consumer market. The predecessor of Apple's commercial operations was "Wang", and the reason was 'graphics', which allowed both human resources and advertising departments to publish inhouse.
~30:00 - My take on the ‘at home or factory question’ is - where is the most value add? As per calculations from Cern, and on this and other channels, a factory can have a bot doing useful work ~21hrs a day. Not many homes would want a bot walking around and trying to find things to do all through the night - in fact it would be kinda spooky. Leasing plans will be tailored to serving the highest need, highest return first. Homes will come in distant second. IMHO
@drknowitallknows Is it actually true that bots will be able to transfer their learnings between one another? I've never seen evidence of learning transfer between models.
So if we're talking about Atlas shrugged. Elon is actually dagny and ilia turns out to be John Galt. Green steel is apparently AI. Can we make saying who is John gault? A thing now
Regarding your speculation on the power consumption of the NVIDIA chip vs. TSLA chip. Would the amount of ventilation holes and cooling provisions on other robots be an indirect way to rank power consumption between robot brands?
Love your videos! 5G communications are installed into watches to make a person's watch act like a super computer even though nearly everything is off loaded to a server. The watch basically becomes just a transmitter. Everything is computed off the watch seeming instantly. I am thinking Tesla bots could use something similar to Xbox Cloud gaming. Using Elon talk, "Delete the computer. The best part is no part." A bot depedant on a continues transmission would be rendered not dangerous as soon as a transmission ends. We would also have a strategy to win the robot apocalypse!
the nVidia GTC keynote has made it clear that the competition in the humanoid robotics sector is fierce. I doubt Tesla has a significant lead over the joint effort of nVidia with the companies integrating nVidia’s GR00T platform. Tesla may achieve a per unit cost advantage due to their vertical integration.
It seems to me that setting up the robotaxi fleet to accept Tesla cars isn't very far from accepting any car driven by an Optimus? And there would sure be lots of advantages to having an Optimus assigned to each vehicle, beyond just driving the vehicle. Help riders with their groceries, luggage, etc. Clean the vehicles as needed between trips and at end of day. Make repairs to the vehicle as needed, certainly change flat tires, but other more difficult repairs should be a piece of cake since Optimus will likely grow up learning all about how the cars are built and set up.
V12.3.4 AUTO speed limit, still can't read overhead LED variable speed limit signs.I know our speed limits are generally too slow on highways but FSD will speed 20+ over the limit which is the 2nd level speeding ticket. Often too fast in residential areas. Counter that with when it slows for corners or vehicles it takes several Kms to begin to speed back up.If I push it back up it is happy to stay there. Doesn't stay right except to pass. When it catches up to a slow truck it slows and waits way too long before passing even if the passing lane is clear, then it slowly accelerates past the truck at a glacial pace. It is way better than V11, smoother, better on mountain hwy corners, slowing. Still slows slightly for some clear green light intersections. Doesn't signal merges (usually).That said it is close to excellent. Before V12 I would use each update a couple times before I gave up out of sheer frustration or annoyance now only the repeated poor weather nag sets me off. Oh, and the auto wipers are still terrible.
This is the good ol' Intel vs ARM debate. The winner is greatest compute at least power. Unfortunately, there's a bit of a trade off. That said, Tesla's inference compute is designed for hard core NN (large dot product) operation at reasonable power consumption.
Added rotation comes at a significant cost and adds failure modes. FIRST PRINCIPLES. How Optimus adds and adapts basic motions and simple skills is totally different from driving in FSD. How Optimus builds subtasks and tasks from basic motions and simple skills is different from FSD. How Optimus navigates through a task is similar to FSD. Most basic motions and simple skills will be stored in Optimus. Common task sequences and local navigation will be stored in Optimus. NEW skills, subtasks, and task sequences will be downloaded from the cloud as needed. Optimus will have at least one AI that FSD does not. That AI will build environment specific tasks from all of the above.
im sure optimus will do simple tasks like picking up and tyre nut and bolting it onto the wheel.... by fall, and more complex tasks, but not complex tasks, by December.
I wonder how long it’ll be before a Gun Holder teaches the Bot to shoot an AR Riffle? Tesla will integrate a no-shooting safeguard in the programme however; can the bot be manipulated? We will see.
Compared to all the other Humanoid Robots, optimus Gen 2 always looks more fluid and ''human' in it's movement.I think this is going to be theme going forward and will set it apart from competent but far less flexible designs.
Any of the AI robots should be able to fine tune their movements by simply monitoring energy expended. Perhaps human movement could be improved on. Perhaps we haven't yet evolved to be as efficient as we could be.
@@mgl455 If there's a need for the T-bot to see behind itself it would be very easy for Tesla to add a rear-facing camera. But if a part is not needed.....
So I hope Tesla using Optimus to build Optimus? I see that have data from as many Optimus robots as possible, like cars are providing millions of hours of training data!
1. The Compute power to train 2. The team of AI experts and every training lesson they have learnt 3. The chips made by Tesla and placed in the cars can be used in the Bots 4. The Tesla factories themselves are traing environments Its hard to say how much FSD knowledge transfers but it will be very high in some areas like compute and methods and much less in others like driving data.
Could you have a high-powered computer near a robot and communicate via Bluetooth or some other 'local' method? It's not an ideal solution but it seems like a possibility.
You seem to be assuming that there will be tasks that require higher compute power than will be built into the standard Optimist. Perhaps. Perhaps we could see some special purpose T-bots that have larger internal computers. But what is likely to be more complex and faster developing than driving a car?
please accept my apologies for not being clear. My point was for competitors who do not have efficient computing. Could they use a local computer that has sufficient capabilities and communicate to a bot nearby? It seems to me this would be a way to get around the energy cost of computation on a bot.
@@donaldpaquet838 I don't think the problem will be in-robot computers. Those appear to be much smaller than what Tesla uses in their cars. The problem, more likely, is having the massive compute available to process the video needed for AI. What I guess will happen, to a large extent, is that companies that have pretty good robots in terms of the mechanics, will pair with companies like Google and Apple who can provide large scale AI development. Tesla has both plus a lot of manufacturing experience.
Voice to text is too big since it makes model too big when you have language detection built in and so on. For example my native language is Estonian and running llm on that is few orders of magnitude harder than most other languages. Estonian language is very hard to learn since word connections and you must have context to decode what are talked about.
Why does all the data Tesla has about driving translate to robots working in a factory? It seems like completely different things. Is it just the navigating an environment part, but learning to turn a screwdriver would need a totally different training dataset?
Re Power budget: FSD HW3 uses almost 100 watts, and HW4 more but less than 200 watts (so 150?). You should not arbitrarily "cut it in half". First FSD now uses both chips in HW3 and is not duplicated in each chip. We do not know about HW4. Unlikely they are building a special version for the bot, I expect the same card in the car as in Optimus. However, I also believe it would be HW5 in Optimus. Why design a brand new product with yesterday's inference engine?
Rather silly points made about the fragility and insecurity of internet connection. It is trivial and relatively cheap to put in a robust and secure network as millions of businesses already do. Plenty of serious points to discuss without scraping the barrel for such silliness
So you were muttering about data protection - and you put your finger on it: in the future everything will be known. Elon was always right: all you need (and get) is TIME. Hail the whatever . . . 😎
Elon is telling the truth! His robots are indeed doing important task at his factories, even as we speak! For example, they are helping the people he just fired to carry all of their personal belongings out of the factories. Think how much money this saves! By the way, as far as Elon promises go, we can expect him to keep all of his promises just like he promised NASA that he would land on the moon with starship in the first three months of this year and of course he did!thank you Elon! We are blessed to have your genius guiding us!
@@civanovici I think as always the ones with the most efficient actuators (maybe with a ReGen of some sort) will be the true winners. When putting down a heavy box, try and regain that power. It's possible,,,, just do it
@goingballistic5 and @DrKnowitall16 have you given any thought to Optimus using 48 volts for the actuators and using the Ethernet ring from Cyber Truck? Seems like those would save weight and simplify wiring
I was talking about AI today and. Long story shortened. I decided we do not have AI. We have AC! Artificial Comprehension. I believe AI is not intelligent because it invents nothing, it just repeats examples. However, I have been very impressed with it reading my C program and understanding it enough to make corrections and ever (It said) make some assumptions. That to me is like Comprehension and it felt as though it was aware (Alive?) But, not intelligent.
I expect GPT-5 to greatly reduce the inference lag when running cloud based models, so I don't see using the cloud for higher brain functions on bots as a big problem in the early years. The Groq chips can already run large models at near real time speeds. By running it off bot you reduce power usage, time to market, and can employ a much more powerful model than running it locally. If lag was a big problem in a factory, you could always set up an on prem server to run the model. I've yet to see Tesla show anything other than a puppet show with Optimus, so concluding that they will dominate the bot market just because their bot has a pretty shell is a stretch.
Headline NEWS (copied from CNBC) on TESLA Auto pilot: Federal regulator finds Tesla Autopilot has 'critical safety gap' linked to hundreds of collisions. Federal authorities say a "critical safety gap" in Tesla's Autopilot system contributed to at least 467 collisions, 13 resulting in fatalities. The findings come from a National Highway Traffic Safety Administration analysis of 956 crashes in which Tesla Autopilot was thought to have been in use. Tesla's Autopilot design has "led to foreseeable misuse and avoidable crashes," The NHTSA said.
I doubt Optimus can catch up with BD, which is 10 years ahead. But that gets to the purpose of the robots. BD is a Terminator - they make no bones about it it is headed for the battlefield, and its main purpose is to kill humans. It is sponsored by DARPA. Optimus doesn't have a stated purpose - that missing White Paper would have defined it. I'm assuming Optimus is designed to take the jobs of tax-paying union jobs. It will be missing critical components that make what it can do very limited it won't have LIDAR, RADR, or 360 degree vision. It will be close to 6 feet tall and weigh 200 pounds with a 4 hour battery that is non-swappable it is a closed system - no 3rd parties bringing thousands of add ons to market Apple's iRobot is designed as an appliance - like an iPhone. My Guess: It won't replace a single human, nor will it hunt and kill humans. It will use LIDAR and RADAR and 360 degree vision to map out the house it serves and after probably 15 minutes it will ID thousands of things in its domain It will be friendly and interact with people - it will be small in normal operation and able to extend "legs" up to clean surfaces and prepare food It will communicate with other iRobots owned by the family at other homes and locations It will fold up to be a small box that weighs probably 30 pounds and is easily transportable to take on vacation and other locations it will have hot-swappable batteries that each power the appliance 24 hours it will be able to climb stairs it will home-school kids it will walk the dog in the backyard It is like an iPhone - with hundreds of thousands of apps by 3rd parties and interchangeable components Apple will sell 1,000X as many of its robots as Elon will sell his Terminator. All because Elon never thought of his robot as a non-threatening appliance but as a replacement for humans another disaster in the making this is so predictable
I drove to Red Lodge and back yesterday on V12 with the menu selection of letting the car decide how fast it goes. I hadn't selected that before and it makes a big positive difference. Coming up to a single lane construction zone with a flagman, it came in a little too hot for me so I braked to a slower speed then put it back in auto. It then slowed nicely to a stop at the flagman where I was first in line. The stop sign flag with slow on the opposite side was on a stand allowing the flagman to not hold on to it. As the opposing line of cars just barely finished coming through, the flagman reached for the flag pole to turn it and even before turning it, the X (very comfortably and not slowly) started to pull around the flagman and go. Just like a human would.
John does a video titled "How 1X Will Beat Tesla!" and a week later he's saying "Tesla’s Optimus Will Completely DOMINATE"
that atlas shrugged joke was too good
I don't get it. -- John Galt.
Thank you guys. It is great to hear two experts discussing this complex topic - instead of a clueless "reporters" who just can't understand the concepts. You have a synergy as friends and colleagues that should be a model for technical content.
I think, Optimus will penetrate the nursing home market before the consumer home. Nursing homes can be equipped with complementary equipment and room layout to accessorize Optimus in a very controlled environment. As accessories become available (aftermarket) so the medical corporations have the $$ to invest. Optimus supervisors (RN’s hopefully) could manage so many more patients, yet provide healthier environments for patients. ER rooms also. An RN would more effectively train the bot.
Cheers from San Diego.
I assume T-bots will flow to where they produce the highest returns for a company. A nursing home is going to gain a lot by having extremely reliable, train-them-once "employees".
The average annual wage for LVNs in California nursing homes is $60,855. It takes between 3 and 3.5 workers to cover a position 24/365. That's over $180k per year plus an additional 30% or so for employer overhead. Perhaps more when calculating in recruitment, training, etc.
The value to a nursing home would be based on avoiding roughly a quarter million dollars per year in staff costs by buying or leasing one T-bot. Few households would be willing/able to outbid a nursing home.
10:20 When it runs out of power... I recall a SciFi story - the locals used steam engines for motive power and imported chips for control. A robot serving drinks started to wobble and the owner shouted to it "Stoke you idiot! Stoke!" and it staggered over to the fireplace, and shoveled a few lumps of anthracite into its firebox to restore steam pressure! Possibilities!
Power sources will be in every room.
Not only is Elon going for the factory market, he is going for his own factories first. Do other manufacturers have anything like that? Any problems Tesla encounters with Optimus in its own factories will be in house problems. Any problems encountered in other factories will NOT be in house problems. There is a world of difference.
Fun video, guys! Much more competition for Tesla in H-bots than self driving, where Tesla's data will be very difficult to beat.
"atlas shrugged" WOW, Love it
@14:41 *Human:* _What do you plan on doing with that knife?_
*Robot:* _I plan to dice the vegetables, slice the potatoes, boil them, cut the bread, sautee the steak, set the table,_
_and if you don't stop pestering me with your asinine questions, I will feed you your own liver with some _*_fava beans,_*_ while I sit back and have some _*_Chianti._*
"Boil 'em, mash 'em, stick 'em in a stew." is what the robot would say about taters
Grok, would for sure.
Scott, that's gotta be the funniest, dad joke. I've heard all week the Highlander.😊❤
An order of magnitude more difficult than training a bot on a task is training a bot how to be part of a team -- a team that may include other bots and/or humans -- where there isn't a 'clean' situation. This is similar to FSD driving on a road by itself (100% solved) versus driving in the real world where stuff happens (99+% solved). This is where Optimus should excel.
Swarm technology? Bots Si, humans No
Robots working in a room or restricted space has the advantage of many cameras watching them. The footage would be realtime processed by a powerful backend AI that does the "digital twin" simulation , thus able to predict with more accuracy each individual bots working. It will relay "overide" commands in realtime when needed to the bots, thus no collision or clashing on the workplace. Those learned model will periodically be incorporated to each bot so the llearned data not from Bot is absorbed.
@@fteoOpty64 That sounds more like the interface planning for KUKA robots. What if there are people in the space? Optimus needs to naviagte the space just like FSD navigates the roads... courteously, carefully, assertively, all in the proper time.
How optimus will roll out..... love it
Good one: "Atlas shrugged"....with leaky hydraulics at that!.
Telsa's FSD goal is level 5. Telsa's Optimus humanoid robot goal is full sentience. This is why Elon wants 25% voting control because his goal is to achieve something that he himself stated has a 10 to 20 percent chance of destroying humanity.
I like the way you think 😎
Not full sentience, that’s ST Data’s goal, to become human.
Optimus goal is to become fully symbiotic with humans.
@@richb2229 In the conference call Elon used the word sentient not symbiotic. " I've said before I think Optimus will be
24:49
more valuable than everything else combined um because if you've got a a
24:55
sentient humanoid robot uh that is able to navigate reality and do tasks at at request um
25:05
there is no meaningful limit to the size of the economy.
Also, I think he'd like more of Tesla because he could have started another company for Optimus, so outside of Tesla.
This is the time we geeks have waited for. All the cool tech and the amazing things that are happening. It's heaven..
v12.3.6 installed today. I haven't even seen a video on it. Intrigued.
“Booster Robotics br002” is the imitation that was referred to in the video by Scott.
Thanks!
Great show guys!!
A G search gives some conflicting info on HW-4 and HW-5. Samsung is the primary producer, but Tesla was messing with TSMC chips. I suspect Samsung is working hard on power usage. I think most Tesla cameras are Samsung. Samsung building a big plant NE of Giga Texas seems important.
In a fixed workspace or cell the Droid operates it could easily be walking on a wireless charging floor.
Scott is priceless 😂
Yes V12.2.4 is so much more polite. I did have issues with yellow green cones twice. It would stop in front of the cones and stop making me look like an idiot and I would have to wave my arms to go around me. This happen twice on the yellow green cone where some workers had painted the street. To get out I had to back up and go around the cones. There was no danger at anytime.
AGREE NUFF SAID!!!!!!!!!!!!!!!!!!!!!!!
Slip rings allow pass through wire connections for joints able to turn 360º
Tesla could use their employees to train Optimus. Elon could say, 'Hey, we want to train Optimus in a home environment. Anyone want to take one home?' Tesla has thousands of employees. How many do these other companies have? A couple hundred at most? There an advantage there.
Tesla - "Take a robot home, and train it."
Employee - "Ok, no problem."
(One week later)
Employee - "Hey, something's wrong. My ID badge doesn't work anymore."
Tesla - "You've been fired. The robot has your job."
Meh... What's few employees more or less? As long as the tests get run and the 'Bots get taught...
I have a problem - my wife fell in love, and ran off with the Optimus bot....
@@charlesrovira5707 Would you say the same thing, if a robot took YOUR job? 🤔
One bot learns, all bots get it. Unlike human, needs to be taught individually.
Yes, the welders definitely put off a lot of high frequency noise. I know this is what I do for a living. Well, now, I'm a T-shirt guy.😊
About privacy thing it is good point. Even when selling only hardware there may be hardware backdoors so that those bots can communicate satelites quite easy and you dont notice anything also you canot verify model as easy as software
Way cool! 😎
10% head count reduction across the company might be making room for Optimus operatives!
Marketing staff is gone. Most likely a large number of people who were writing FSD code and doing labeling are gone.
When the bots are moving 10x faster and more fluidly and they put those bots on the assembly line to make more bots... You'll have Botception. The wait until they start Gigapressing bots. Holy crap
Atlas shrugged 😂😅🤪🤣🤣🤣
I would like to see you both explain in layman’s terms the meaning of inverse kinematics as this knowledge is key robot movement and articulation
You could do voice to model directly using wavetovec embedding. However, most LLM are trained on text and not wave (voice)
If your typing's as bad as mine. John try jet GBT 3.5. It's free the Highlander.😊
It seems like zero turn wheels would be a lot more efficient than legs for factory work.
battery life may not mean so much, when many work stations will allow you to plug the thing into a cord? Also, thanks for the cool show.
I'm not as worried about Optimus destroying humanity as about humanity destroying itself. In my lengthy lifetime we have come perilously close to doing that on our own. We have lived under a sword of Damocles for so many years we have come to accept it as unlikely, rather than inevitable.
Optimus may instead save us from ourselves, or at least remain behind to tell our legend.
Hey John sitting and listening to your video and thought... If Tesla want to get voice to text, then all employees could get a microphone for training and relevant phrases
Just curious, why would the humanoid bot use 1 chip when the car uses 2? What is the reason for using 2 chips in the car, is it for redundancy? Excellent video!! Starting it off with that joke put a smile to my face
Apple was already leveraged in the education and financial markets before it ramped into the consumer market. The predecessor of Apple's commercial operations was "Wang", and the reason was 'graphics', which allowed both human resources and advertising departments to publish inhouse.
Not coming home anytime soon as 100LB bot falling on your dog would be bad, and you can only make it so light.
JOHN AND SCOTT HAVE GREAT CHEMISTRY.....
GREAT VIDEOS.....
LINE OF THE VIDEO: "ATLAS SHRUGGED!!!" - SCOTT
~30:00 - My take on the ‘at home or factory question’ is - where is the most value add? As per calculations from Cern, and on this and other channels, a factory can have a bot doing useful work ~21hrs a day. Not many homes would want a bot walking around and trying to find things to do all through the night - in fact it would be kinda spooky. Leasing plans will be tailored to serving the highest need, highest return first. Homes will come in distant second. IMHO
@drknowitallknows Is it actually true that bots will be able to transfer their learnings between one another? I've never seen evidence of learning transfer between models.
So if we're talking about Atlas shrugged. Elon is actually dagny and ilia turns out to be John Galt. Green steel is apparently AI.
Can we make saying who is John gault? A thing now
Regarding your speculation on the power consumption of the NVIDIA chip vs. TSLA chip. Would the amount of ventilation holes and cooling provisions on other robots be an indirect way to rank power consumption between robot brands?
Tesla should use police body cam footage for training Optimus!
Love your videos! 5G communications are installed into watches to make a person's watch act like a super computer even though nearly everything is off loaded to a server. The watch basically becomes just a transmitter. Everything is computed off the watch seeming instantly. I am thinking Tesla bots could use something similar to Xbox Cloud gaming. Using Elon talk, "Delete the computer. The best part is no part." A bot depedant on a continues transmission would be rendered not dangerous as soon as a transmission ends. We would also have a strategy to win the robot apocalypse!
Aiming for the home market isn’t such a big risk. If it doesn’t work out they can pivot to business instantly
So about 15 W with the human brain. 100 wattouch for the total human body, that still puts us ahead of the robot.😊
Just keep Optimus away from the pool or
the nVidia GTC keynote has made it clear that the competition in the humanoid robotics sector is fierce. I doubt Tesla has a significant lead over the joint effort of nVidia with the companies integrating nVidia’s GR00T platform. Tesla may achieve a per unit cost advantage due to their vertical integration.
At some point the bot will be called: Optimus Prime. Dojo will create the All Spark. And it will more than meets the eye
It seems to me that setting up the robotaxi fleet to accept Tesla cars isn't very far from accepting any car driven by an Optimus? And there would sure be lots of advantages to having an Optimus assigned to each vehicle, beyond just driving the vehicle. Help riders with their groceries, luggage, etc. Clean the vehicles as needed between trips and at end of day. Make repairs to the vehicle as needed, certainly change flat tires, but other more difficult repairs should be a piece of cake since Optimus will likely grow up learning all about how the cars are built and set up.
When do you guys believe we will be able to start selling compute time of our cars to Tesla and possibly others?
I wonder how many of us being amazed at this technology will suddenly find there are no jobs
V12.3.4 AUTO speed limit, still can't read overhead LED variable speed limit signs.I know our speed limits are generally too slow on highways but FSD will speed 20+ over the limit which is the 2nd level speeding ticket. Often too fast in residential areas. Counter that with when it slows for corners or vehicles it takes several Kms to begin to speed back up.If I push it back up it is happy to stay there. Doesn't stay right except to pass. When it catches up to a slow truck it slows and waits way too long before passing even if the passing lane is clear, then it slowly accelerates past the truck at a glacial pace. It is way better than V11, smoother, better on mountain hwy corners, slowing. Still slows slightly for some clear green light intersections. Doesn't signal merges (usually).That said it is close to excellent. Before V12 I would use each update a couple times before I gave up out of sheer frustration or annoyance now only the repeated poor weather nag sets me off. Oh, and the auto wipers are still terrible.
This is the good ol' Intel vs ARM debate. The winner is greatest compute at least power. Unfortunately, there's a bit of a trade off. That said, Tesla's inference compute is designed for hard core NN (large dot product) operation at reasonable power consumption.
I can see why FSD has a large moat. I don't see the large moat for the bot.
Added rotation comes at a significant cost and adds failure modes. FIRST PRINCIPLES.
How Optimus adds and adapts basic motions and simple skills is totally different from driving in FSD. How Optimus builds subtasks and tasks from basic motions and simple skills is different from FSD. How Optimus navigates through a task is similar to FSD. Most basic motions and simple skills will be stored in Optimus. Common task sequences and local navigation will be stored in Optimus. NEW skills, subtasks, and task sequences will be downloaded from the cloud as needed. Optimus will have at least one AI that FSD does not. That AI will build environment specific tasks from all of the above.
im sure optimus will do simple tasks like picking up and tyre nut and bolting it onto the wheel.... by fall, and more complex tasks, but not complex tasks, by December.
I wonder how long it’ll be before a Gun Holder teaches the Bot to shoot an AR Riffle? Tesla will integrate a no-shooting safeguard in the programme however; can the bot be manipulated? We will see.
Compared to all the other Humanoid Robots, optimus Gen 2 always looks more fluid and ''human' in it's movement.I think this is going to be theme going forward and will set it apart from competent but far less flexible designs.
Any of the AI robots should be able to fine tune their movements by simply monitoring energy expended. Perhaps human movement could be improved on. Perhaps we haven't yet evolved to be as efficient as we could be.
Ya gotta admit though, mechanically the new atlas bot is extremely cool
I prefer the new atlas with 360 degree vision. Why should it have to rotate it’s head?
@@mgl455
If there's a need for the T-bot to see behind itself it would be very easy for Tesla to add a rear-facing camera. But if a part is not needed.....
So I hope Tesla using Optimus to build Optimus?
I see that have data from as many Optimus robots as possible, like cars are providing millions of hours of training data!
How possible is it to use fsd in Optimus?
Completely it’s AI so extremely flexible.
1. The Compute power to train
2. The team of AI experts and every training lesson they have learnt
3. The chips made by Tesla and placed in the cars can be used in the Bots
4. The Tesla factories themselves are traing environments
Its hard to say how much FSD knowledge transfers but it will be very high in some areas like compute and methods and much less in others like driving data.
Could you have a high-powered computer near a robot and communicate via Bluetooth or some other 'local' method? It's not an ideal solution but it seems like a possibility.
You seem to be assuming that there will be tasks that require higher compute power than will be built into the standard Optimist.
Perhaps. Perhaps we could see some special purpose T-bots that have larger internal computers. But what is likely to be more complex and faster developing than driving a car?
please accept my apologies for not being clear.
My point was for competitors who do not have efficient computing. Could they use a local computer that has sufficient capabilities and communicate to a bot nearby? It seems to me this would be a way to get around the energy cost of computation on a bot.
@@donaldpaquet838
I don't think the problem will be in-robot computers. Those appear to be much smaller than what Tesla uses in their cars. The problem, more likely, is having the massive compute available to process the video needed for AI.
What I guess will happen, to a large extent, is that companies that have pretty good robots in terms of the mechanics, will pair with companies like Google and Apple who can provide large scale AI development.
Tesla has both plus a lot of manufacturing experience.
Don t they obviously need a bigger model for the bot then for the car?
Surely bots raised in the Tesla environment will grow up to believe anything is possible and will be finished in two weeks?
Voice to text is too big since it makes model too big when you have language detection built in and so on. For example my native language is Estonian and running llm on that is few orders of magnitude harder than most other languages. Estonian language is very hard to learn since word connections and you must have context to decode what are talked about.
Why does all the data Tesla has about driving translate to robots working in a factory? It seems like completely different things. Is it just the navigating an environment part, but learning to turn a screwdriver would need a totally different training dataset?
Tesla cars already take voice commands. Is that speech to text?
Re Power budget: FSD HW3 uses almost 100 watts, and HW4 more but less than 200 watts (so 150?). You should not arbitrarily "cut it in half". First FSD now uses both chips in HW3 and is not duplicated in each chip. We do not know about HW4. Unlikely they are building a special version for the bot, I expect the same card in the car as in Optimus. However, I also believe it would be HW5 in Optimus. Why design a brand new product with yesterday's inference engine?
👍
Rather silly points made about the fragility and insecurity of internet connection. It is trivial and relatively cheap to put in a robust and secure network as millions of businesses already do. Plenty of serious points to discuss without scraping the barrel for such silliness
Imagine someone successfully doing their job with a tele-operated humanoid. Those are the jobs it will be able to do, practically none!
Only a 12 year old wants a robot.... there is no practical use for a human form robot.
So you were muttering about data protection - and you put your finger on it: in the future everything will be known. Elon was always right: all you need (and get) is TIME. Hail the whatever . . . 😎
I could see the collapse of the Hollywood movie industry. And then you can just see it coming.😊
How come nobody's talking about MKBHD short with multiple Optimi in his workplace?
Likely because it wasn't real.
'wireless charging' is required for optimus and tesla cars.
Do you think Optimus will ever be
able to tend my garden? Is it waterproof?
I think it will be do any task a human can do ANY and jobs/tasks we can’t do also
Elon is telling the truth! His robots are indeed doing important task at his factories, even as we speak! For example, they are helping the people he just fired to carry all of their personal belongings out of the factories. Think how much money this saves! By the way, as far as Elon promises go, we can expect him to keep all of his promises just like he promised NASA that he would land on the moon with starship in the first three months of this year and of course he did!thank you Elon! We are blessed to have your genius guiding us!
Ima Mexican Mama ready for the mopping and where’s my Optimus Robot?! 🤨🦾
How long until Human/Robo marriage legal?
It walks like Biden!
The Nvidia 'Jetson' inference modual only requires about 10 watts to function.
it depends, jetson orin AGX between 15-75W
@@civanovici I think as always the ones with the most efficient actuators (maybe with a ReGen of some sort) will be the true winners. When putting down a heavy box, try and regain that power. It's possible,,,, just do it
@goingballistic5 and @DrKnowitall16 have you given any thought to Optimus using 48 volts for the actuators and using the Ethernet ring from Cyber Truck? Seems like those would save weight and simplify wiring
Is protecting AI intellectual property similar to protecting bitcoin transactions? Jeannine
Home bots will need a configuration setting like FSD "chill / average / assertive" for how anal you are about silverware or t-shirts.
I was talking about AI today and.
Long story shortened. I decided we do not have AI.
We have AC! Artificial Comprehension. I believe AI
is not intelligent because it invents nothing, it just repeats
examples.
However, I have been very impressed with it reading my
C program and understanding it enough to make corrections
and ever (It said) make some assumptions.
That to me is like Comprehension and it felt as though it
was aware (Alive?) But, not intelligent.
They should lease Optimus, not sell.
RoboRobot ! I'm a robot .. click every palm tree, every bicycle, every bus....Zzzzzzzz.
Who is Elon Musk.
?
Love your enthusiasm, however, please try not to talk over each other.
Brett Adcock slug-a-glugg'ed Sunny D as a kid. you can tell. Right from the bottle.
What?
I expect GPT-5 to greatly reduce the inference lag when running cloud based models, so I don't see using the cloud for higher brain functions on bots as a big problem in the early years. The Groq chips can already run large models at near real time speeds. By running it off bot you reduce power usage, time to market, and can employ a much more powerful model than running it locally. If lag was a big problem in a factory, you could always set up an on prem server to run the model. I've yet to see Tesla show anything other than a puppet show with Optimus, so concluding that they will dominate the bot market just because their bot has a pretty shell is a stretch.
I agree this a BULL STRETCH. IT IS ALL SPECULATION!😮
Headline NEWS (copied from CNBC) on
TESLA Auto pilot: Federal regulator finds
Tesla Autopilot has 'critical safety gap' linked to hundreds of collisions. Federal authorities say a "critical safety gap" in Tesla's Autopilot system contributed to at least 467 collisions,
13 resulting in fatalities.
The findings come from a National Highway Traffic Safety Administration analysis of 956 crashes in which Tesla Autopilot was thought to have been in use.
Tesla's Autopilot design has "led to foreseeable misuse and avoidable crashes," The NHTSA said.
I doubt Optimus can catch up with BD, which is 10 years ahead.
But that gets to the purpose of the robots.
BD is a Terminator - they make no bones about it
it is headed for the battlefield, and its main purpose is to kill humans.
It is sponsored by DARPA.
Optimus doesn't have a stated purpose - that missing White Paper would have defined it.
I'm assuming Optimus is designed to take the jobs of tax-paying union jobs.
It will be missing critical components that make what it can do very limited
it won't have LIDAR, RADR, or 360 degree vision.
It will be close to 6 feet tall and weigh 200 pounds with a 4 hour battery that is non-swappable
it is a closed system - no 3rd parties bringing thousands of add ons to market
Apple's iRobot is designed as an appliance - like an iPhone.
My Guess:
It won't replace a single human, nor will it hunt and kill humans.
It will use LIDAR and RADAR and 360 degree vision to map out the house it serves
and after probably 15 minutes it will ID thousands of things in its domain
It will be friendly and interact with people - it will be small in normal operation
and able to extend "legs" up to clean surfaces and prepare food
It will communicate with other iRobots owned by the family at other homes and locations
It will fold up to be a small box that weighs probably 30 pounds and is easily transportable
to take on vacation and other locations
it will have hot-swappable batteries that each power the appliance 24 hours
it will be able to climb stairs
it will home-school kids
it will walk the dog in the backyard
It is like an iPhone - with hundreds of thousands of apps by 3rd parties and interchangeable components
Apple will sell 1,000X as many of its robots as Elon will sell his Terminator.
All because Elon never thought of his robot as a non-threatening appliance
but as a replacement for humans
another disaster in the making
this is so predictable
Spoken with a lot of missing brain cells.
WTH is dis dude spitting 😂