If things are really going to progress faster than Moore's law, that's ironic because people thought Moore's law would stop because progress would slow down.
I grew up with computers and have noticed these changes in perspective in last 30 years: - in the 90's most videos had 24Hz if I remember correctly and all studies was saying everything above that is worthless since our eyes can't process more. Highest number I currently saw was like 360hz and it's always a difference. - for longer than a decade they were saying (look up the science) that we will never be able to go for processors to build under 7nm. We passed that about 10 years ago and we kept going. With AGI is the same we keep moving the goal post to admit is here. Our minds can't really comprehend exponentiality. I believe not a single person would have guessed how far we got just this year, for someone that got on a personal level on the AI wave, and I was expecting growth I am amazed of everything I can do by myself, how much I can learn, how much my brain processing power and patterns have changed just this year by interacting with AI on daily basis.
Moore's law hasn't been accelerated though, as it states that transistor density at minimum cost levels would increase exponentially, doubling about every 2 years. Increasing your globally available compute by increasing other factors such as increasing the total manufacturing volume, shrinking precision of operations (to do more operations with fewer transistors) is a great way to increase certain goals, but it doesn't invalidate anything about what we have seen about Moore's law itself (that it hasn't kept up, especially on the cost front).
Hi David, what you do is truly invaluable! You are one of the best commentators on the field of ai and how rapidly it has been growing! we all value the human voice and touch you provide to the discussion! No Ai can take that away! Keep doing what you’re doing!
In addition to the factors you mentioned (such as: robots are cheaper than cars) it's possible that robot production will scale in a unique way compared to other tech, because robots can assist in building more robots. Even if only 10% of robots produced are reinvested into production (ie, used to help build robots), that could scale overall robot production pretty quickly (assuming no shortages in any materials/components). Robots might achieve ubiquity surprisingly quickly. Fingers crossed...
Exactly! We might only be able to build 1.7B robots the first year, but when those robots turn around and build more robots, our capacity goes up exponentially. Cars that come out of car factories can’t turn around and build more cars.
Robots are a fairly difficult hardware task, but not an impossible one. Battery technology is pretty much there to make them useful for extended periods of time. The difficulty is in the software. I fully believe the cost of a humanoid robot can be in the realm of 20-30k in todays money. Perhaps less if made at a truly massive industrial scale. That would be a true revolution. Elon calls humanoid robots "the biggest business opportunity, ever" and I agree.
@@coolbanana165 Why would your employer pay your robot when they can buy/rent their own robot for a fraction of the cost? Maybe if Government(s) force employers to only use Robots/AI that are owned/rented from real Humans? Seems possible, but unlikely?
Regarding robot production, the robots can assemble the robots. Factory lines with classical industrial robots might still be more efficient for some steps of production. But in principle, one robot can create another, two make 4, 4 make 8, 8 make 16 … (after 30 iterations you end up at a billion) Right now production is mostly “bottlenecked” by the steps for which you need humans, as the production step requires a degree of flexibility that industrial (not general purpose) robots do not have. This limitation could be quickly overcome with general purpose robots by the robot workforce scaling up the robot work force (not limited to robots, it would also allow the scaling up of car manufacturing, or hardware manufacturing of any kind)
What you are not considering is that the first clients of humanoids are probably going to be the companies that manufacture them, and they are going to be used both to fabricate more robots and to make more factories, that will accelerate manufacturing 10-100 X and will drive the cost per unit close to the cost of the raw materials. So within a couple of years we could have a enough robots to kick off the singularity
I myself had similar thoughts. Although I suspected double exponential growth. Maybe overkill but current progress is definitely faster than standard exponential function.
And we have not seen recursive self improvement yet, at least not big scale. Once we have AGI or early ASI (like organization level intelligence, OpenAIs level 5, essentially the equivalent of all of NVIDIA, Microsoft or intel) in AI, development speed might increase. Automation really makes a difference once the human is out of the loop, till then there will always be a bottleneck, slowing things down
Think about it like this it takes about 25 years to create a highly skilled human being. And God only knows how much money. We will be able to create smarter, harder working, and as many as we want, robots. You're not going to be worried about the price of these. This is going to flip the economy upside down but not in a bad way. It will make everyone rich.
You are making a bold claim about capitalism, which tends always to make the rich richer and to keep the poor poor. Will need TONS of very convincing evidence. Not holding breath.
@@patpowers9210 I believe that capitalism will cease to exist in roughly 7 to 10 years. As the world economic forum has stated for quite some time. In the future you will own nothing, and you will be happy.
ASICs make a lot of sense in principle, also as they can be more performant by around a factor of 1000x, but the process of building them is expensive, otherwise we would make ASICs for a lot more things then we do right now. If the process of creating ASICs is however increasingly automated thanks to transformers, ASIC-builders (ASIC- creation ASICs ;-) AGI, etc. we might do it a lot more, as the threshold to get to break even drops. Also in everything AI at the same time the demand will increase
Great comparison from cars to building robot humanoids and that we already have the capacity to build them, just need to allocate resources appropriately. I am really looking forward to seeing how humanity's opportunities to live and enjoy life open up so much more when AI-first economies of scale roll into place.
Using cars as a scalability metric was tangible presentation for me and greatly assisted my understanding of what to expect in the future. The second hand car market is another aspect of things to come for bots as they have a pretty long service life. Crazy respect for your field of interest/passion and willingness to share expertise with others 🇦🇺
Here in “meat space” I’m dealing with the reality of a relative in the medical system. A MRI session cost $20K, compared to ~ $10K in 2017. The doctors and nurses at the hospital are using technology reminiscent of the turn of the century. I see zero evidence of any ML or AI at any level. This large hospital’s billing system is medieval. Moores Law is “squared”? Sorry, but I see no evidence of that in the day-day.
I recently retired from corrections and have similar experiences / observations. Although “technology” (laptops, cellphones) were incorporated over my nearly 20 year career, our inefficiency and incompetence was (purposely) mind blowing. Our online tools were not intuitive and virtually unmanageable. In 2024, we still had two fax machines: one for “outgoing” faxes & one for “incoming” faxes. Humanoid robots will be sending and receiving faxes in my office 50 years from now. The govt cannot adapt.
I worry this video exemplifies a category of thinking in terms of what we "can" do rather than what we "should" do. The only argument I heard was one of economic efficiency, as if there is some sort of inherent morality in magnifying economic output and/or profit. Such thinking led to obesity crisis, rust belt, predatory lending and healthcare, and other issues which are still in progress and hitherto minimally mitigated. We've pushed most lifeforms on this planet to the fringes, and in the process lost countless beautiful and fragile species. This also is an ongoing issue which shows no signs of slowing or stabilizing. Soon we may be the next species that will get squashed and pushed to the side, as we thoughtlessly race ahead without any idea of what we are trying to achieve.
It is so interesting that we are living in the future. It is natural now to have earbuds, endless entertainment possibilities, supercomputers in our pocket... And these are getting cheaper and cheaper. So in a few years it will be natural to have multifunctional robots at home. Just like mobile phones, internet. In the 90-s, only few people had it and in 10 years everybody have one. I know it's not a big deal... I'm just practicing the English language. 😅
This humanoid would possess value far exceeding the sum of its individual components, offering capabilities that surpass those of any traditional support worker. With advanced artificial intelligence and robotics, it could seamlessly integrate various functionalities, such as emotional support, task management, and real-time problem-solving. Unlike human support workers, who may face limitations due to fatigue or emotional stress, this android would be designed to operate tirelessly and consistently, providing unwavering support. Its ability to learn and adapt over time would further enhance its effectiveness, allowing it to cater to specific needs and preferences with remarkable precision. This transformative technology could revolutionize the way support is delivered, creating a level of care and assistance that is not only more efficient but also profoundly impactful on the lives of those it serves.
I think one problem is that you've basically got to cram a data center into a robot which is probably not possible space wise. Getting rid of the heat and providing it with enough power is another problem. As you note we are close to transistor shrinkage limitations (due to physics) and the extra computing power is just coming from increasing the number of cards in the data centers. There is talk of AI companies building their own nuclear power plants to provide their data centers with sufficient power. So while human level AGI may be around the corner, cramming the computing power needed into a robot may not be (you could perhaps control the robot remotely from the data center but you are probably not doing that for hundreds of thousands of robots).
50% of the value each robot generates needs to be taxed and that tax revenue used for UBI. There should be an exemption for "homestead" robots, probably one or two per household. Yes it's complicated but that's the core of it. We probably also need a compute tax for machine intelligence.
We need to tax resource usage, not creation of value. Tax land usage with a land value tax, CO2 emissions, pulling resources out of the ground etc. That should be the first line of taxes. Every other tax should be carefully tacked on only if really needed and for controlling the behavior of the actors in the economy.
It's not going to be about money. You can have the robots build inexpensive housing for humans. You can use AI to create inexpensive entertainment for humans. You can use robots to raise inexpensive food for humans. And you can use robots to give inexpensive caretaking to humans and if you use older medicines you can provide inexpensive Medicare to humans. Ubi might not be in the form of cash so much as a place to live with food and entertainment. And then if you can figure out some kind of a gig for additional income you can climb out of that class of people. The biggest problem is going to be wealthy and powerful people trying to block the robots who could easily replace them.
If we get these massive productivity boosts, two things will happen. The government will swallow an even bigger chunk of the economy and everything will get much more affordable and cheaper.
If millions of people lose their jobs due to humanoid robots then they wont be able to buy cars so resources will get transferred to robots production and there will be fewer cars. This will probably also encourage greater use of autonomous vehicles which will be cheaper
Cars also don't usually built themselves. If you build the robots and they can build robots then those robots can build more robots to build more robots. Sad part is no one will pay us to do a job that a robot can do faster and or cheaper. We are obsoleting ourselves very quickly.
I think every human should be allocated a robot which they can employ anyway they deem necessary. People don't go to work anymore. They just look online to see what company needs a robot and they send a robot off to do the job for them. Thats how we will make a little bit of extra money.
Houses, cars, speedboats, and similar assets generally retain their monetary value better than computer parts and computers. Homes and vehicles can appreciate over time due to factors like location, demand, and market trends. In contrast, computer components often experience rapid depreciation, primarily driven by the fast-paced nature of technological advancements, which quickly render older models obsolete. A well-maintained home or a reliable car can not only hold its value but may also appreciate, making them more stable long-term investments compared to the rapidly evolving landscape of computer technology. Additionally, items such as paper, pencils, and books have also seen a decrease in value, reflecting broader market trends, particularly with the rise of digital alternatives.
David, big fan of yours. Good luck on the novel. I wrote a book a network protocols that took me 12 years. I'd rather eat a Buick than do it again ;-) BTW, check out the sci fi story by Jack Williamson called "With Folded Hands. About welcoming our robot overlords. Cheers!
Absolutely want to put the brains in the robots, NOT in the data centers. This is my best use case for my Photocore design. We need unhackable robots with high intelligence, high autonomy, and high efficiency. The quantum dots in my Photocore design are so efficient they can emit or recieve single photons or single electrons.
Just give me AGI, lets get it done. Remove conventional jobs. Introduce UBI. I want to live in a cabin in the wilderness with my friend. Roam the lands
Are people really talking about using ASICs? They are really more for mass production as you can't program them on the fly. I would have thought FPGAs would be the go-to choice for the moment (although they don't have the electronic flexibility of a dedicated ASIC).
To stick the dismount I think the top concern is the priority list of which human roles get displaced first, second, third... Downstream from certain labor displacements, if groups of humans are unable to participate in the economy, it'll cause extreme need. What if we focus on autonomous robotic production of (1) water (2) energy (3) food -- codified in law (with fangs). Robots managing our water & sewage. Robots sowing/growing/nurturing/protecting/reaping crops, packaging, loading, transporting (trains/depots/transfers), unloading at endpoints, distributing (brick-n-mortar stores, drone deliveries). Autonomous supply chain. Focus on the prime enablers in law, so that they are not negotiable. If these things aren't comprehensively in place, everything pauses until they are. With these fundamentals guaranteed, the capsizing-capable force of rebellion can be avoided. If we don't address how to roll this out in a certain order, we're going to have a lot of friction. Enough to spoil the fun.
I'd be interested to listen to a discussion about the following: Assuming that human median problem solving (100 IQ) "System 2 Thinking" is equivalent to GPUs/TPUs that can handle changing the state of 100 Terabytes in 1/30 of a second, 2024 is the first year that a data center has reached this threshold. What are your thoughts?
Every feature you want eill cost you extra guaranteed. Wash the dishes extra $15 per day. Do some advanced molecular research extra $500 etc... The corporations will find ways to bleed everyone for every dollar.
You guys ever noticed how when corporations talk about robots and AI they talk about how they will make trillions of dollars but when they talk about robots for you it's always going to make your bed and make your life easier but nothing about how your going to make money if they paid you what your worth you wouldn't need AI or robots you could just get a maid
I can't even get the best frontier LLMs to do my software engineering job for me reliably. They help a lot, but they still need about 50% help on everything and often times go off in wild, erroneous directions and need to be painstakingly reset or put back on track. It's hard to imagine solving the much, much more complex problems of navigating the real world to do even simple tasks, like loading the dishwasher and running it, reliably in the next few years. They will need to do things that are as-yet unsolved, such as complex planning, persistent and useful remembering of things, innate understanding of physics, they will need to be able to actually learn. For robots to be a thing, they need not only a new and different AI computing paradigm from LLMs, which hasn't been invented yet, but also at least an order of magnitude more advanced general capabilities. Lots of advancements are converging, to be sure, but we're still very far away from house robots.
Plus you can build robots to help build more robots. e.g. robots that build more manufacturing facilities. robots that increase mining (note: rio tinto mining is heavily in automation). robots can increase almost all levels of production, thus increasing the growth of production. (That said, I am a bit terrified of the effects to the environment. note: rio tinto mining.)
I would bet we’ll choose the specialization/modular design. Most things don’t really need a model to rule them all, sounds like a waste of energy. I could be wrong though, it is possible a complete model for anything and everything could end up being very efficient though it sounds risky
so the robot does everything, while you are walking the woods all day every day, enjoying the beach , shopping, with money you no longer have a job, to earn money from? sounds like a Wall-e dystopia!
I wouldn't say that we will converge into a one model to rule the world situation. From a technological standpoint there is no one single computer to rule the world there are multiple manufacturers and each operating system is different in its own way. So, with that being said I doubt seriously that there will come a time when all the models converge into one.
Do some introspection. When (say) walking up stairs with a drink in your hand look at it and wonder what part of you is doing it, because it's not "you"
There's even better use for introspection - trying to find "you" in you. Consensus is strong over the history in some (mostly eastern) traditions that it's not really anywhere and "you" is just illusion and not exactly very useful one, btw.
Sharp guys often say that Moore's Law has to slow down. Here is a sharp guy saying it will speed up because of what he thinks are underrated factors. 🤖
I would like to know your take on generative AI in the universities and how higher education will be impacted longer term by all the level 2 thinking agents.
I would like to think universities remain important, but developing scientific oriented AI research bots will make learning material more and more removed from human comprehension.
its not going to be 900/m to lease you said its 50k and 10k/year for maintenance at 900 is just enough to pay for maintenance not pay for the the principal or make a profit
Moores law... The most abused statistic ever created lol Original observation was that they could shrink the transistor density to half every ~18 months. After the initial early days of transistor gains were done, and instructions got larger and more effective, it transitioned to computers being roughly 2x as fast every 18 months. And as that petered out, the defenition changed to performance per watt doubling every 18 months... Which was the stupidest and shortest lived Intel definition of Moores law ever lol. When you get outside of Intel, you can pick your company or techno-philosopher, and everyone picks their own definition to suit their purposes to claim that "something is roughly 2x as good in a little less than 2 years" Point is... Moores law is a buzz word like synergy, or block chain. Sure, there are legit things behind those buzz words, but 99% of the time when a ceo says them it has more to do with stock manipulation than any real thought or program. There is too much noise, and too little definition to the term to be useful, and any ceo claiming it is coming up on a hard end of a trend rather than continuing it.
A robot that can give you a piggyback home from the pub could replace a car.
This man is living in the future
You made me chuckle.
Just imagining a drunk office worker hopping on the back of an atlas robot shouting "hi-ho silver!"
As a British man, I agree
How about a virtual pub that is so hyper realistic that you can just stay home and drink with your friends?
If things are really going to progress faster than Moore's law, that's ironic because people thought Moore's law would stop because progress would slow down.
Moore's Hypothesis
I grew up with computers and have noticed these changes in perspective in last 30 years:
- in the 90's most videos had 24Hz if I remember correctly and all studies was saying everything above that is worthless since our eyes can't process more. Highest number I currently saw was like 360hz and it's always a difference.
- for longer than a decade they were saying (look up the science) that we will never be able to go for processors to build under 7nm. We passed that about 10 years ago and we kept going. With AGI is the same we keep moving the goal post to admit is here. Our minds can't really comprehend exponentiality. I believe not a single person would have guessed how far we got just this year, for someone that got on a personal level on the AI wave, and I was expecting growth I am amazed of everything I can do by myself, how much I can learn, how much my brain processing power and patterns have changed just this year by interacting with AI on daily basis.
Moore's law hasn't been accelerated though, as it states that transistor density at minimum cost levels would increase exponentially, doubling about every 2 years.
Increasing your globally available compute by increasing other factors such as increasing the total manufacturing volume, shrinking precision of operations (to do more operations with fewer transistors) is a great way to increase certain goals, but it doesn't invalidate anything about what we have seen about Moore's law itself (that it hasn't kept up, especially on the cost front).
For close to a decade it was a sure way to get clicks.
It’s Moores law, ASICs, Software breakthroughs, economy of scale (as we produce more hardware), money flowing into the sector… all put together.
Yeah so when I looked at the data behind what he said, I was actually kinda surprised TBH.
Hi David, what you do is truly invaluable! You are one of the best commentators on the field of ai and how rapidly it has been growing! we all value the human voice and touch you provide to the discussion! No Ai can take that away! Keep doing what you’re doing!
A video that _really_ _could_ use visuals to visualise the information but instead is a single image.
In addition to the factors you mentioned (such as: robots are cheaper than cars) it's possible that robot production will scale in a unique way compared to other tech, because robots can assist in building more robots. Even if only 10% of robots produced are reinvested into production (ie, used to help build robots), that could scale overall robot production pretty quickly (assuming no shortages in any materials/components). Robots might achieve ubiquity surprisingly quickly. Fingers crossed...
Exactly! We might only be able to build 1.7B robots the first year, but when those robots turn around and build more robots, our capacity goes up exponentially. Cars that come out of car factories can’t turn around and build more cars.
Robots are a fairly difficult hardware task, but not an impossible one. Battery technology is pretty much there to make them useful for extended periods of time. The difficulty is in the software. I fully believe the cost of a humanoid robot can be in the realm of 20-30k in todays money. Perhaps less if made at a truly massive industrial scale. That would be a true revolution. Elon calls humanoid robots "the biggest business opportunity, ever" and I agree.
@@Shrouded_reaper Unitree's humanoid robot is only $16k. Can definitely go way lower.
@@Shrouded_reaperin a domestic setting battery life is less important like having a laptop plugged in permanently but is still useful.
My robot will get repo'd when I lose my job to a robot.
This.
Send your robot to take your job.
Profit
@@coolbanana165 My previous employer already bought them my position is no longer available.
@@coolbanana165 Why would your employer pay your robot when they can buy/rent their own robot for a fraction of the cost?
Maybe if Government(s) force employers to only use Robots/AI that are owned/rented from real Humans? Seems possible, but unlikely?
If you could afford your own robot, you could rent it out and send it to work. It'd end up paying for itself.
Damn, that's a catchy title. I clicked that notification so fast lmao
@@lesliejohnrichardson That word combo Though.. 😉
Same here
Regarding robot production, the robots can assemble the robots. Factory lines with classical industrial robots might still be more efficient for some steps of production. But in principle, one robot can create another, two make 4, 4 make 8, 8 make 16 … (after 30 iterations you end up at a billion) Right now production is mostly “bottlenecked” by the steps for which you need humans, as the production step requires a degree of flexibility that industrial (not general purpose) robots do not have. This limitation could be quickly overcome with general purpose robots by the robot workforce scaling up the robot work force (not limited to robots, it would also allow the scaling up of car manufacturing, or hardware manufacturing of any kind)
We should be grateful that it's Moore's law and not Murphy's law
Or Matthew's law
What you are not considering is that the first clients of humanoids are probably going to be the companies that manufacture them, and they are going to be used both to fabricate more robots and to make more factories, that will accelerate manufacturing 10-100 X and will drive the cost per unit close to the cost of the raw materials. So within a couple of years we could have a enough robots to kick off the singularity
Jensen Huang says what he has to say for Nvidia stock prices to keep rising.
Hello stochastic parrot!
Yes indeed, talking his book to a room full of sales weasels. We’re theeeese close to AGI… 😂
The original Moore's law was also stated by Gordon Moore (founder of intel) to raise stock prices of the company he worked at ;)
Ace! Thanks mate. Something good to watch before bed (South Korea here)
More like to listen.
I myself had similar thoughts. Although I suspected double exponential growth. Maybe overkill but current progress is definitely faster than standard exponential function.
And we have not seen recursive self improvement yet, at least not big scale. Once we have AGI or early ASI (like organization level intelligence, OpenAIs level 5, essentially the equivalent of all of NVIDIA, Microsoft or intel) in AI, development speed might increase. Automation really makes a difference once the human is out of the loop, till then there will always be a bottleneck, slowing things down
Think about it like this it takes about 25 years to create a highly skilled human being. And God only knows how much money. We will be able to create smarter, harder working, and as many as we want, robots. You're not going to be worried about the price of these. This is going to flip the economy upside down but not in a bad way. It will make everyone rich.
You are making a bold claim about capitalism, which tends always to make the rich richer and to keep the poor poor. Will need TONS of very convincing evidence. Not holding breath.
@@patpowers9210 I believe that capitalism will cease to exist in roughly 7 to 10 years. As the world economic forum has stated for quite some time. In the future you will own nothing, and you will be happy.
@@patpowers9210that‘s American capitalism. In Europe it‘s fine.
ASICs make a lot of sense in principle, also as they can be more performant by around a factor of 1000x, but the process of building them is expensive, otherwise we would make ASICs for a lot more things then we do right now. If the process of creating ASICs is however increasingly automated thanks to transformers, ASIC-builders (ASIC- creation ASICs ;-) AGI, etc. we might do it a lot more, as the threshold to get to break even drops. Also in everything AI at the same time the demand will increase
The materials cost for an ASIC chip is less than a dollar. We pay for the technology.
Great comparison from cars to building robot humanoids and that we already have the capacity to build them, just need to allocate resources appropriately. I am really looking forward to seeing how humanity's opportunities to live and enjoy life open up so much more when AI-first economies of scale roll into place.
Using cars as a scalability metric was tangible presentation for me and greatly assisted my understanding of what to expect in the future.
The second hand car market is another aspect of things to come for bots as they have a pretty long service life. Crazy respect for your field of interest/passion and willingness to share expertise with others 🇦🇺
Sounds about right. Supplying energy to so many robots could be a problem, but with ASI, it may be solved quickly as well.
I feel like there was a PowerPoint with this that did not make it to UA-cam.
Here in “meat space” I’m dealing with the reality of a relative in the medical system. A MRI session cost $20K, compared to ~ $10K in 2017. The doctors and nurses at the hospital are using technology reminiscent of the turn of the century. I see zero evidence of any ML or AI at any level. This large hospital’s billing system is medieval.
Moores Law is “squared”? Sorry, but I see no evidence of that in the day-day.
I recently retired from corrections and have similar experiences / observations. Although “technology” (laptops, cellphones) were incorporated over my nearly 20 year career, our inefficiency and incompetence was (purposely) mind blowing. Our online tools were not intuitive and virtually unmanageable. In 2024, we still had two fax machines: one for “outgoing” faxes & one for “incoming” faxes.
Humanoid robots will be sending and receiving faxes in my office 50 years from now. The govt cannot adapt.
he's been saying something similar for a while now, and he's not wrong. (but yes, this is the first time i've heard him say "moore's law squared")
Its more like Moore's Law to the nth power. It's both variable and unknown; it could very well be much higher or lower than two at any given time.
I worry this video exemplifies a category of thinking in terms of what we "can" do rather than what we "should" do. The only argument I heard was one of economic efficiency, as if there is some sort of inherent morality in magnifying economic output and/or profit. Such thinking led to obesity crisis, rust belt, predatory lending and healthcare, and other issues which are still in progress and hitherto minimally mitigated.
We've pushed most lifeforms on this planet to the fringes, and in the process lost countless beautiful and fragile species. This also is an ongoing issue which shows no signs of slowing or stabilizing. Soon we may be the next species that will get squashed and pushed to the side, as we thoughtlessly race ahead without any idea of what we are trying to achieve.
It is so interesting that we are living in the future. It is natural now to have earbuds, endless entertainment possibilities, supercomputers in our pocket... And these are getting cheaper and cheaper. So in a few years it will be natural to have multifunctional robots at home. Just like mobile phones, internet. In the 90-s, only few people had it and in 10 years everybody have one. I know it's not a big deal... I'm just practicing the English language. 😅
Ah...Carvana..Now I am thinking a vending machine of robots.
This humanoid would possess value far exceeding the sum of its individual components, offering capabilities that surpass those of any traditional support worker. With advanced artificial intelligence and robotics, it could seamlessly integrate various functionalities, such as emotional support, task management, and real-time problem-solving. Unlike human support workers, who may face limitations due to fatigue or emotional stress, this android would be designed to operate tirelessly and consistently, providing unwavering support. Its ability to learn and adapt over time would further enhance its effectiveness, allowing it to cater to specific needs and preferences with remarkable precision. This transformative technology could revolutionize the way support is delivered, creating a level of care and assistance that is not only more efficient but also profoundly impactful on the lives of those it serves.
A worker robot is the perfect asset. You can even take depreciation credit on it 😂
No need to own car if my robot can haul a rickshaw around. Can even rent it out like an Uber.
I think one problem is that you've basically got to cram a data center into a robot which is probably not possible space wise. Getting rid of the heat and providing it with enough power is another problem. As you note we are close to transistor shrinkage limitations (due to physics) and the extra computing power is just coming from increasing the number of cards in the data centers. There is talk of AI companies building their own nuclear power plants to provide their data centers with sufficient power. So while human level AGI may be around the corner, cramming the computing power needed into a robot may not be (you could perhaps control the robot remotely from the data center but you are probably not doing that for hundreds of thousands of robots).
50% of the value each robot generates needs to be taxed and that tax revenue used for UBI. There should be an exemption for "homestead" robots, probably one or two per household. Yes it's complicated but that's the core of it. We probably also need a compute tax for machine intelligence.
We need to tax resource usage, not creation of value. Tax land usage with a land value tax, CO2 emissions, pulling resources out of the ground etc. That should be the first line of taxes. Every other tax should be carefully tacked on only if really needed and for controlling the behavior of the actors in the economy.
It's not going to be about money.
You can have the robots build inexpensive housing for humans.
You can use AI to create inexpensive entertainment for humans.
You can use robots to raise inexpensive food for humans.
And you can use robots to give inexpensive caretaking to humans and if you use older medicines you can provide inexpensive Medicare to humans.
Ubi might not be in the form of cash so much as a place to live with food and entertainment. And then if you can figure out some kind of a gig for additional income you can climb out of that class of people.
The biggest problem is going to be wealthy and powerful people trying to block the robots who could easily replace them.
Robot tax collector with machine guns to force you to pay
I bought the book last night (B&N), waiting for it to arrive in a few days.
With all this and UBI, what is your stock tip?
If we get these massive productivity boosts, two things will happen. The government will swallow an even bigger chunk of the economy and everything will get much more affordable and cheaper.
And that $900 a month boils down to $1.67 per hour (assuming 25% downtime for charging and maintenance)…
Oh good, can’t wait for Used Robot Salesmen to become a thing!
If millions of people lose their jobs due to humanoid robots then they wont be able to buy cars so resources will get transferred to robots production and there will be fewer cars. This will probably also encourage greater use of autonomous vehicles which will be cheaper
With a decent enough robot you could share it with neighbors.
Cars also don't usually built themselves. If you build the robots and they can build robots then those robots can build more robots to build more robots. Sad part is no one will pay us to do a job that a robot can do faster and or cheaper. We are obsoleting ourselves very quickly.
I think every human should be allocated a robot which they can employ anyway they deem necessary. People don't go to work anymore. They just look online to see what company needs a robot and they send a robot off to do the job for them. Thats how we will make a little bit of extra money.
Houses, cars, speedboats, and similar assets generally retain their monetary value better than computer parts and computers. Homes and vehicles can appreciate over time due to factors like location, demand, and market trends. In contrast, computer components often experience rapid depreciation, primarily driven by the fast-paced nature of technological advancements, which quickly render older models obsolete. A well-maintained home or a reliable car can not only hold its value but may also appreciate, making them more stable long-term investments compared to the rapidly evolving landscape of computer technology. Additionally, items such as paper, pencils, and books have also seen a decrease in value, reflecting broader market trends, particularly with the rise of digital alternatives.
The world has no idea whats just around the corner
David, big fan of yours. Good luck on the novel. I wrote a book a network protocols that took me 12 years. I'd rather eat a Buick than do it again ;-) BTW, check out the sci fi story by Jack Williamson called "With Folded Hands. About welcoming our robot overlords. Cheers!
Absolutely want to put the brains in the robots, NOT in the data centers. This is my best use case for my Photocore design. We need unhackable robots with high intelligence, high autonomy, and high efficiency. The quantum dots in my Photocore design are so efficient they can emit or recieve single photons or single electrons.
i 💚 Jensen, the Godfather of ai
Just give me AGI, lets get it done. Remove conventional jobs. Introduce UBI. I want to live in a cabin in the wilderness with my friend. Roam the lands
why the wilderness? Why not a beautiful cyberpunk aesthetic big city?
@@quantumspark343I want you to think about this.
@@tomizatko3138 With AI in not-so-nice hands, they can and will be able to make you think it.
@@tomizatko3138 a cyberpunk city is more comfortable and aesthetically pleasing, i dont want the dystopian part either obviously, just the tech part
Moore Law squared ... at the cost of price cubed.
Let us know when the the audiobook for Heavy Silver comes out!
You think you will ever cover Keen Technologies? John Carmack's AGI company?
If it’s so smart how is it not a person? If it is a person how can you justify owning one?
Easy it is not a person. It is my robot.
I will give it the personality I want and it will serve me until I buy it's replacement.
@@historicallyintriguing-q2p watch meet the robinsons
My computer is so smart and it is not a person.
Are people really talking about using ASICs? They are really more for mass production as you can't program them on the fly. I would have thought FPGAs would be the go-to choice for the moment (although they don't have the electronic flexibility of a dedicated ASIC).
To stick the dismount I think the top concern is the priority list of which human roles get displaced first, second, third... Downstream from certain labor displacements, if groups of humans are unable to participate in the economy, it'll cause extreme need. What if we focus on autonomous robotic production of (1) water (2) energy (3) food -- codified in law (with fangs). Robots managing our water & sewage. Robots sowing/growing/nurturing/protecting/reaping crops, packaging, loading, transporting (trains/depots/transfers), unloading at endpoints, distributing (brick-n-mortar stores, drone deliveries). Autonomous supply chain.
Focus on the prime enablers in law, so that they are not negotiable. If these things aren't comprehensively in place, everything pauses until they are. With these fundamentals guaranteed, the capsizing-capable force of rebellion can be avoided.
If we don't address how to roll this out in a certain order, we're going to have a lot of friction. Enough to spoil the fun.
UBI can cover a good deal this problem.
That is funny. An orderly transition will not happen. It's called disruption for a reason.
@@greatcondor8678 Agreed, of course. We're seeing a train wreck in slow motion. Just saying the obvious for the sake of it.
I'd be interested to listen to a discussion about the following:
Assuming that human median problem solving (100 IQ) "System 2 Thinking" is equivalent to GPUs/TPUs that can handle changing the state of 100 Terabytes in 1/30 of a second, 2024 is the first year that a data center has reached this threshold. What are your thoughts?
Every feature you want eill cost you extra guaranteed. Wash the dishes extra $15 per day. Do some advanced molecular research extra $500 etc... The corporations will find ways to bleed everyone for every dollar.
I GOTTA TRY THAT ASIC IDEA!!
You guys ever noticed how when corporations talk about robots and AI they talk about how they will make trillions of dollars but when they talk about robots for you it's always going to make your bed and make your life easier but nothing about how your going to make money if they paid you what your worth you wouldn't need AI or robots you could just get a maid
I can't even get the best frontier LLMs to do my software engineering job for me reliably. They help a lot, but they still need about 50% help on everything and often times go off in wild, erroneous directions and need to be painstakingly reset or put back on track. It's hard to imagine solving the much, much more complex problems of navigating the real world to do even simple tasks, like loading the dishwasher and running it, reliably in the next few years. They will need to do things that are as-yet unsolved, such as complex planning, persistent and useful remembering of things, innate understanding of physics, they will need to be able to actually learn. For robots to be a thing, they need not only a new and different AI computing paradigm from LLMs, which hasn't been invented yet, but also at least an order of magnitude more advanced general capabilities. Lots of advancements are converging, to be sure, but we're still very far away from house robots.
Plus you can build robots to help build more robots. e.g. robots that build more manufacturing facilities. robots that increase mining (note: rio tinto mining is heavily in automation). robots can increase almost all levels of production, thus increasing the growth of production. (That said, I am a bit terrified of the effects to the environment. note: rio tinto mining.)
I thought it was hyperbole until i reread the title and saw the quote source 😅
Yeah, one look at the size of the dies on Blackwell tells us all we need to know about "Mooke's Law" squared. 🙄
Notwithstanding Huang’s credentials, it’s in his best interests to hype anything that will sell more product. Interesting thoughts, nonetheless.
I would bet we’ll choose the specialization/modular design. Most things don’t really need a model to rule them all, sounds like a waste of energy.
I could be wrong though, it is possible a complete model for anything and everything could end up being very efficient though it sounds risky
exponent squared is exponential at twice the rate (9 months)
That was a great video.
Would you let the robot do the UA-cam broadcasts?
It will be robots doing the UA-cam videos, and it will be robots watching those videos. Economy solved.
I’ve always thought exponential x exponential is the singularity.
Fei Fei world labs is working on spatial models
Imagine taking your car to the dealership and not being charged $240 an hour..
all that you say, sounds Moore or less correct ;)
so the robot does everything, while you are walking the woods all day every day, enjoying the beach , shopping, with money you no longer have a job, to earn money from? sounds like a Wall-e dystopia!
900 a month? Yes, I'd do that. Paying for occasional home assistance is more expensive than that.
I wouldn't say that we will converge into a one model to rule the world situation. From a technological standpoint there is no one single computer to rule the world there are multiple manufacturers and each operating system is different in its own way. So, with that being said I doubt seriously that there will come a time when all the models converge into one.
Am calling this the gundam theory.
Where video?
I feel sorry for early adopters of robots. A robot from 2026 will seem like a model T by 2029 at 1/3 cost.
Do some introspection. When (say) walking up stairs with a drink in your hand look at it and wonder what part of you is doing it, because it's not "you"
There's even better use for introspection - trying to find "you" in you. Consensus is strong over the history in some (mostly eastern) traditions that it's not really anywhere and "you" is just illusion and not exactly very useful one, btw.
Moore’s lore
Moore's Flaw.
In the end who will buy goods and services, most people won't have an income?
When do you think we will have Mars colony?
No, we're seeing Wirth's law squared.
Moore's Law squared is ALSO a Moore's Law. It is an exponential, dammit!
it is Moore's law with 2x speed. Which is very significant.
"Robot, go fix my car".. Don't you get "The Orville" vibes?
Having experienced the cruelty of kids, even toward an adult, yeah.
I get the nexus comment like fear monger and stuff you have to accept he make some interesting points about AI.,and the way we function
I wonder, will tech devolopment keep going even after the soceatal and İnstitutional collapse due to this + private ai and ai sciencetist tech???
Sharp guys often say that Moore's Law has to slow down. Here is a sharp guy saying it will speed up because of what he thinks are underrated factors. 🤖
I would like to know your take on generative AI in the universities and how higher education will be impacted longer term by all the level 2 thinking agents.
I would like to think universities remain important, but developing scientific oriented AI research bots will make learning material more and more removed from human comprehension.
MAID ANDROID GIRLS WILL RISE!
its not going to be 900/m to lease
you said its 50k and 10k/year for maintenance at 900 is just enough to pay for maintenance not pay for the the principal or make a profit
llets the martrix storyline begin.
If it is sentient is it a piece of equipment?
Should it called Jensen Huang law now since he owns the world basically?
Moores law was doomed from the beginning to become less more's law.
Didn't Moore's law say that the price also drops?
Yeah i dont know why he is leaving that part out. This is the whole point of moor's law.
Do you have a list of your current Github project that we can participate in?
Plus many new robots will begin building other robots, it will be exponential.
I understand that in the Netherlands robots in tge automotive sector pay union fees
First video with AI Voiceover?
never has technology moved faster than moores law? moores law itself hasn't even been around that long compared to technology in general lol
Moores law... The most abused statistic ever created lol
Original observation was that they could shrink the transistor density to half every ~18 months.
After the initial early days of transistor gains were done, and instructions got larger and more effective, it transitioned to computers being roughly 2x as fast every 18 months. And as that petered out, the defenition changed to performance per watt doubling every 18 months... Which was the stupidest and shortest lived Intel definition of Moores law ever lol.
When you get outside of Intel, you can pick your company or techno-philosopher, and everyone picks their own definition to suit their purposes to claim that "something is roughly 2x as good in a little less than 2 years"
Point is... Moores law is a buzz word like synergy, or block chain. Sure, there are legit things behind those buzz words, but 99% of the time when a ceo says them it has more to do with stock manipulation than any real thought or program. There is too much noise, and too little definition to the term to be useful, and any ceo claiming it is coming up on a hard end of a trend rather than continuing it.
A lot of what I hear in AI is hype with some really important advances mixed in here and there.
♥️