Why Tesla’s Optimus Will Surprise (Ep. 751)

hey it's Dave so a couple days ago I put out a video about how we can expect a billion humanoid robots in 20 years or so and John Gibs from the channel Dr know- itall he did a reaction video and he had some really thoughtful things to say I'll link it in the video description but it got me thinking more about Optimus uh humanoid robot from a few different angles overall the point of this video will be to talk about how I think Tesla's Optimus humanoid robot will probably surprise people at how fast it becomes useful in fact I wouldn't be surprised if we get a use humanoid robot in the next one to two years anyways I could be wrong but here's my thinking first let's make a guess on where Tesla stands in their effort to make a useful humanoid robot let's say a humanoid robot is basically the body and the brains so two main parts the body is the robots Hardware so including all of the actuators right the the body the electronics Etc and then the brain is the neuron Nets that control the robot autonomously to do useful things so how far along is Tesla with the body and the brain so I'm not talking about I say a final refined version of the robot but rather how far are they along toward a first version of a truly useful and competent humanoid robot I think with the body I'd say maybe they're about 70% maybe 80% I'm sure they have a ton of things left to refine and improve the body but I think they've got the major parts done well they've integrated their own aurator design into the robot they've completed a very compelling design of fingers and hands and the walking is coming along too I think a lot of the hardware seems like it's in place now how about the brain for the brain I think they're probably only about 10 to 20% there let's say I think it really hasn't been a focus of the team yet but I think the focus has probably already shifted to the brain and when you look at the video of Optimus folding laundry we can see Tesla is already starting to focus on training the robot with Advanced tasks like folding clothes one advantage of Tesla's approach is they've been working on FSD for so long that they've created a realworld foundation AI model and this model is going to be basically the bones that help the car but also humanoid robots understand and interact with the physical world around them I think 2024 probably will be the year where Tesla Optimus makes a huge jump in its brain capability so second why is this exciting largely because you don't need a perfect robot to deploy it or even to sell it this is in stark contrast to FSD now FSD or full self driving is probably one of the most challenging use cases of AI it's literally dealing with life and death situations at almost every move so with FSD you don't want to release a version that claims that people don't need to supervise it unless it truly delivers Stellar performance it needs to be close to perfect in terms of not causing accidents but contrast that with a humanoid robot a humanoid robot is not dealing with safety critical things most of the time probably it would be rarely used right for safety critical tasks at least in the early versions of the robot take the example of some of the restaurant robots that I've seen lately some robots are more like moving carts that hold Beverages and deliver them to tables these things just sense their surroundings and navigate to the right table but because they're moving so slow they're not going to kill someone and that's the advantage that robots have over driving in driving on the road you can't make mistakes and you can't just move super slow to avoid things if you move too slow you can actually cause accidents another example is chat GPT chat GPT had its breakthrough moment because it didn't need to be perfect it could make mistakes and hallucinate but it still was useful and that's because it's not dealing with life and death situations most of the time in a similar way a humanoid robot can be deployed even when it's not perfect it can stutter it could be a bit slow hey it can even take like 5 Seconds to think about something right before doing it or it can refuse to do something altogether if it doesn't know how to do it so this means the risk of deploying a humanoid robot is a lot less than the risk of deploying an unsupervised FSD version to vehicles and it's probably one of the biggest reasons I think we'll see rapid Improvement and progress with a humanoid robot now cha GPT is improving much faster by deploying to millions of people because all of that usage is going back into training the model to make it better I think with the humano robot I think we could see something similar rather than releasing a perfect robot Tesla just needs to reach a point where the robot can safely deliver compelling value and if the value is high enough then the demand will be there and then you can leverage the user base to train better AI models and increase the value the robot is able to give the humanoid robot is basically going to have practically endless room for improvement and as it improves the value it gives increases thus it increases its demand so I asked CH GPT list out let's say 20 things that a humano robot will be able to do that can give value to people and here's a few things that it came up with one is house cleaning right laundry services Pet Care like walking your dog you have Fitness coaching um home security and monitoring I that would be fantastic you have gardening and Landscape Elder Care Child Care Assistance cooking and meal prep running errands I think this list is really just a drop in the bucket because if you think about it the ultimate potential of the human robot lies in its ability to learn new things on the Fly the old Paradigm of robots was that you had to program the robot to do a task and the robot did that task but the new paradigm of robots opened up by neuron Nets is this idea that a robot can adapt and learn new tasks without even specifically being programmed to do that task in other words a robot that can learn now of course there will be limitations especially early on the robot's Hardware might be kind of clunky and slow the AI neuron Nets might not be able to pick up every complex task just by looking at how they're done but there's a certain level of Competency that once the robot reaches then it can do a lot of useful things if a humanoid robot can put laundry in the laundry machine put it in the dryer take it out fold the clothes and then put the clothes away in your drawer now that's helpful How about if a human robot can take a vacuum cleaner and vacuum the whole house or how about the robot taking your dog out for a walk not just once but once in the morning and once at night maybe even taking your dog out to go potty before your dog sleeps hey even getting your dog's food in its bowl and getting your dog more water right in its water bowl these things might sound trivial but they are really hard tasks for robots to do right now and it's mainly a hardware and an AI issue but with the right hardware and neuron Nets if you can solve these types of problems then it unlocks a whole host of other things that the robot can do and again we're not talking about a static robot that can only do what it's programmed to do rather this is a robot that's designed from the ground up to be constantly learning how to do new tasks can you imagine the overthe a software updates for Optimus now that's going to be awesome to experience as an Optimus robot owner the robot just gets a lot smarter overnight with new capabilities anyways there's lots of exciting potential for Optimist and I think Tesla is doing an amazing job here and is potentially unlocking a massive Market anyways on a personal note I'm here in Vancouver British Columbia this month my wife is doing some natural treatment to try to prevent recurrence of cancer for those following my wife found two breast cancer tumors over the summer last year she ended up having surgery a couple months after to remove them and so far we haven't found evidence that the cancer has spread so that's good news but what we're trying to do is to take proactive action and do what we can do to prevent recurrence so we've changed a lot of our diet and exercise habits and we've been seeking out additional treatments to help so thanks for everyone who've kept my wife and my family in your thoughts it's deeply appreciated we'll see you guys in my next video thanks

Channel: Dave Lee

Views: 75,136

Keywords:

Id: 1VlvDcfN12U

Length: 7min 53sec (473 seconds)

Published: Sat Jan 20 2024