Barry Crisp

This will change the world - forever. On August 19th, 2021, Tesla unveiled the Optimus Bot. Well... unveiled is probably a little bit generous. They showed a person in a robot costume doing some sick dance moves, followed by a 10-minute presentation outlining their plans to expand their product lineup to humanoid robots. Fast forward 2 years - Tesla has multiple prototypes available that are capable of walking on their own, pick up and put down objects, navigate their surroundings, and execute tasks like sorting. There's a link from Tesla in the replies below that shows the latest version sorting objects by color. At a glance, this may not be very impressive, especially when you compare it to Boston Dynamic's robots doing backflips and parkour. But how it learned to sort is the breakthrough I want to focus on, which has mind-blowing implications not just for Tesla, but the global labor market. Videos in, controls out. This is a theme that @elonmusk - Tesla's CEO - has been talking about for quite a while. The premise is to build a Neural-Net system (fancy word for AI) that doesn't require humans to explicitly write code that tells a machine what to do. If you've been following my posts, you may be aware of Tesla achieving a self-driving version for their EVs that is run entirely using this process (FSD v12). They have an AI brain at their HQ that analyzes a tremendous amount of video data collected by their cars, and then the AI brain tells the car how to navigate in every scenario it faces on the road. Instead of a human writing code to account for stop signs, traffic lights, etc., Tesla has an AI that has learned how to do this by observing what driving looks like. This is a really big deal. What this means is that Tesla is now constrained by how many video clips it can gather from its EV fleet, and how many chips they have (both from NVIDIA and their in-house DOJO chip) to process that data. They are no longer limited by a breakthrough in "code" - they have an AI brain that can figure this out with enough examples. What's more - this method of solving for real-world driving can apply to any physical task. All you need is video in, and the AI will spit controls out. Enter the Tesla Optimus Bot. Believe it or, even though the Bot pictured below and a Tesla car look like 2 completely different objects, they share a lot more in common than it appears. They are both physical objects that use software to navigate their environments. They both use the same on-board computer to process said software. They both use the same batteries to power the motors that allow each object to move around. And they both use the AI-brain that teaches itself how to perform tasks by analyzing countless video clips. In the video that I've linked in the replies, you will see the Bot sorting blocks into their respective areas by their color. Based on the information shared by Tesla thus far, it's safe to assume the Bot was able to do this not because of code that was written by a human that says "pick up blue block, put in blue area"... But by analyzing video clips of blocks being sorted by their appropriate color - no different than how their cars learn to drive themselves. This is highlighted by a maneuver that may not seem like much, but demonstrates how powerful this approach is. In the video linked below, starting at the 45 second mark, you will see the Bot self-correcting a block that was sorted on its side. What this likely means is that the AI-brain had video clips of the objects being sorted right-side up, instead of on their sides. The bot automatically, without the need of human code, understood that the block it sorted landed on its side, picked it back up, adjusted the orientation, and placed it back down on the correct side. This means that the bot is able to dynamically adjust without any explicit instructions on how to handle the complexities of the real world. As long as it has an end result (sort the blocks by color), it will deal with all the little nuances that our 3D world can throw at you. The amount of code that a human would need to write to account for this is mind-blowing. Instead, Tesla's AI-brain has figured out how to handle it by analyzing video. You can see another example of this 34 seconds in where a person randomly moves around blocks on the table, yet the Bot is able to track the objects on its own without the need of explicit instructions from a human. The implications of this are ridiculous. As long as Tesla is able to manufacture a robot that can reliably execute commands from a physical perspective - meaning that the actuators, batteries, hands, joins, etc. are built to be extremely durable and capable of repeatedly handling tasks - and Tesla is able to manufacture this robot at a reasonable price... The world will never be the same. With enough strength and dexterity, Tesla's Bot could handle almost all physical tasks by simply looking at video clips of people doing said tasks. Picking up a vacuum and running it through the house. Sorting and folding laundry. Tidying up the house. Moving material from point A to point B. Picking up trash and placing it in a bin. Pushing a lawnmower. Monitor an area for safety-related concerns. Laying bricks. Hammering nails. Using power tools. Clean dishes... etcetera, etcetera, etcetera. Like its cars, the Bot is not limited by some sort of breakthrough in code to handle said tasks. It's limited by how many video clips and chips Tesla can process for its AI-brain to tell the Bot what to do. However - there are many steps that still need to be solved. Tesla needs to invent and manufacture actuators, sensors, and materials that are capable to reliably handle any and all situations the Bot could be used for, and they need to manufacture these at extremely low costs so that this technology is affordable. Luckily for the company, one of its advantages is manufacturing, being the only automaker in the world that is capable of manufacturing EVs at an affordable price and do so profitably. Tesla's Model Y is the bestselling car in the world so far in 2023, while making the most profit per car out of any other mass-market automaker regardless of gas, hybrid, or electric. Their track record so far seems to back up this claim. And now, with their Bot, Tesla is starting to transition into a product category that a huge majority of the world thinks is decades, if not millennia away. But in reality, the company is knocking on the door of a paradigm shift that could up-end disrupting what it means to have a job. Let's run through a simple exercise that drives this point home. When looking at the raw materials involved for this robot, the costs don't appear to be astronomically high. It's unlikely Tesla will use too many exotic materials to build the robot - perhaps actuators which are the little motors that will operate the robot's limbs. Outside of that, you've got cameras, wires, a computer that Tesla already builds en masse for its cars, rubbers, etc. In addition, Elon Musk has previously said that the cost of the robot should not exceed the cost of your average car, so that many people could afford it. This lines up when looking at the bot from a materials perspective. Under this scenario, let's assume the price of the Bot will land somewhere between $25k and $50k to the consumer after accounting for the cost of manufacturing at full ramp. Now let's take an excerpt from @WalterIsaacson's biography of Elon Musk that was released on 9/12. In this book, there's an excerpt from a discussion between Elon and his engineers where the goal for the bot should be to operate for 16 hours without a charge. That is equal to 2 8-hour shifts' worth of human labor, completely uninterrupted. Now let's take your standard job that requires physical labor - I will let you decide which jobs you think this bot would be able to accomplish and come up with an hourly rate for said jobs. For my example, I'll use maid, averaging around $15 an hour. If a maid works full time (40 hours per week) for a year, they will bring home $31,200 in the first year, and double that in the second year ($62,400). As you can see, in our use case, the upper-end cost of a Bot at $50k is already lower than the 2-year cost of hiring a maid. But here's the deal - the bot is designed to work 16-hour shifts, which means 80-hour weeks. This is because bots don't need to rest, sleep, and go on strike. They're not human. That means that in year one, the Bot will be able to do double the work of a maid, while costing 20% less. And here's where it gets really nutty - starting in year two, the cost of the Bot is $0. That's because you've already paid $50k up front. Instead of paying a maid $32k every year, for let's say 5 years, which equals to $160k over 5 years - you pay $50k up front for the bot, and you get DOUBLE THE WORK in those 5 years because it can work 16-hour shifts. After 5 years, you would've paid the equivalent of $320k for a maid to do the same amount of work a Tesla Bot did for $50k. That's a reduction of 85% in labor costs over 5 years. There may be monthly subscription fees that Tesla could charge in order to do certain tasks - but these will likely pale in comparison to how much one would spend on the human-equivalent of each task. If Tesla is able to manufacture millions of these per year, and they're capable of doing most physical human tasks, why would anyone pay a human to do said tasks? This does two things - it dramatically lowers the cost of labor to a point where products and services could be a fraction of the cost vs where they are now, and it gives businesses no reason to hire a human for 7x the costs over 5 years to do the same exact job. The savings get larger the longer the Bot can work for without replacement. Oh and by the way, the employer won't need to pay the bot any health insurance, 401k, and it will never be concerned about any HR risks. The reality is that this future is much closer than many realize. It appears that Tesla has solved for the most difficult problem in human labor - an AI-brain that will automatically generate actions based on video that it analyzes in the real world. And with its manufacturing expertise, they should be able to crank out millions of these per year over the course of the next couple decades that should bring massive abundance but could also create massive chaos. The robots are coming. As a society, we need to begin thinking about the future that we want with a technology like this existing in our daily lives. More importantly, we need to put in safeguards so that this technology doesn't cause more harm than it does good. We want a world of plenty for everyone, not a world of plenty for the few. Executed correctly, this technology can enable a future that few of us thought possible.

farzyness's tweet photo. This will change the world - forever.

On August 19th, 2021, Tesla unveiled the Optimus Bot. Well... unveiled is probably a little bit generous.

They showed a person in a robot costume doing some sick dance moves, followed by a 10-minute presentation outlining their plans to expand their product lineup to humanoid robots.

Fast forward 2 years - Tesla has multiple prototypes available that are capable of walking on their own, pick up and put down objects, navigate their surroundings, and execute tasks like sorting.

There's a link from Tesla in the replies below that shows the latest version sorting objects by color. At a glance, this may not be very impressive, especially when you compare it to Boston Dynamic's robots doing backflips and parkour.

But how it learned to sort is the breakthrough I want to focus on, which has mind-blowing implications not just for Tesla, but the global labor market.

Videos in, controls out.

This is a theme that @elonmusk - Tesla's CEO - has been talking about for quite a while. The premise is to build a Neural-Net system (fancy word for AI) that doesn't require humans to explicitly write code that tells a machine what to do.

If you've been following my posts, you may be aware of Tesla achieving a self-driving version for their EVs that is run entirely using this process (FSD v12). They have an AI brain at their HQ that analyzes a tremendous amount of video data collected by their cars, and then the AI brain tells the car how to navigate in every scenario it faces on the road.

Instead of a human writing code to account for stop signs, traffic lights, etc., Tesla has an AI that has learned how to do this by observing what driving looks like.

This is a really big deal.

What this means is that Tesla is now constrained by how many video clips it can gather from its EV fleet, and how many chips they have (both from NVIDIA and their in-house DOJO chip) to process that data. They are no longer limited by a breakthrough in "code" - they have an AI brain that can figure this out with enough examples.

What's more - this method of solving for real-world driving can apply to any physical task. All you need is video in, and the AI will spit controls out.

Enter the Tesla Optimus Bot.

Believe it or, even though the Bot pictured below and a Tesla car look like 2 completely different objects, they share a lot more in common than it appears.

They are both physical objects that use software to navigate their environments. They both use the same on-board computer to process said software. They both use the same batteries to power the motors that allow each object to move around. And they both use the AI-brain that teaches itself how to perform tasks by analyzing countless video clips.

In the video that I've linked in the replies, you will see the Bot sorting blocks into their respective areas by their color. Based on the information shared by Tesla thus far, it's safe to assume the Bot was able to do this not because of code that was written by a human that says "pick up blue block, put in blue area"...

But by analyzing video clips of blocks being sorted by their appropriate color - no different than how their cars learn to drive themselves.

This is highlighted by a maneuver that may not seem like much, but demonstrates how powerful this approach is. In the video linked below, starting at the 45 second mark, you will see the Bot self-correcting a block that was sorted on its side.

What this likely means is that the AI-brain had video clips of the objects being sorted right-side up, instead of on their sides.

The bot automatically, without the need of human code, understood that the block it sorted landed on its side, picked it back up, adjusted the orientation, and placed it back down on the correct side.

This means that the bot is able to dynamically adjust without any explicit instructions on how to handle the complexities of the real world. As long as it has an end result (sort the blocks by color), it will deal with all the little nuances that our 3D world can throw at you.

The amount of code that a human would need to write to account for this is mind-blowing. Instead, Tesla's AI-brain has figured out how to handle it by analyzing video. You can see another example of this 34 seconds in where a person randomly moves around blocks on the table, yet the Bot is able to track the objects on its own without the need of explicit instructions from a human.

The implications of this are ridiculous.

As long as Tesla is able to manufacture a robot that can reliably execute commands from a physical perspective - meaning that the actuators, batteries, hands, joins, etc. are built to be extremely durable and capable of repeatedly handling tasks - and Tesla is able to manufacture this robot at a reasonable price...

The world will never be the same.

With enough strength and dexterity, Tesla's Bot could handle almost all physical tasks by simply looking at video clips of people doing said tasks.

Picking up a vacuum and running it through the house. Sorting and folding laundry. Tidying up the house. Moving material from point A to point B. Picking up trash and placing it in a bin. Pushing a lawnmower. Monitor an area for safety-related concerns. Laying bricks. Hammering nails. Using power tools. Clean dishes... etcetera, etcetera, etcetera.

Like its cars, the Bot is not limited by some sort of breakthrough in code to handle said tasks. It's limited by how many video clips and chips Tesla can process for its AI-brain to tell the Bot what to do.

However - there are many steps that still need to be solved. Tesla needs to invent and manufacture actuators, sensors, and materials that are capable to reliably handle any and all situations the Bot could be used for, and they need to manufacture these at extremely low costs so that this technology is affordable.

Luckily for the company, one of its advantages is manufacturing, being the only automaker in the world that is capable of manufacturing EVs at an affordable price and do so profitably.

Tesla's Model Y is the bestselling car in the world so far in 2023, while making the most profit per car out of any other mass-market automaker regardless of gas, hybrid, or electric. Their track record so far seems to back up this claim.

And now, with their Bot, Tesla is starting to transition into a product category that a huge majority of the world thinks is decades, if not millennia away. But in reality, the company is knocking on the door of a paradigm shift that could up-end disrupting what it means to have a job.

Let's run through a simple exercise that drives this point home.

When looking at the raw materials involved for this robot, the costs don't appear to be astronomically high. It's unlikely Tesla will use too many exotic materials to build the robot - perhaps actuators which are the little motors that will operate the robot's limbs.

Outside of that, you've got cameras, wires, a computer that Tesla already builds en masse for its cars, rubbers, etc. In addition, Elon Musk has previously said that the cost of the robot should not exceed the cost of your average car, so that many people could afford it. This lines up when looking at the bot from a materials perspective.

Under this scenario, let's assume the price of the Bot will land somewhere between $25k and $50k to the consumer after accounting for the cost of manufacturing at full ramp.

Now let's take an excerpt from @WalterIsaacson's biography of Elon Musk that was released on 9/12. In this book, there's an excerpt from a discussion between Elon and his engineers where the goal for the bot should be to operate for 16 hours without a charge.

That is equal to 2 8-hour shifts' worth of human labor, completely uninterrupted.

Now let's take your standard job that requires physical labor - I will let you decide which jobs you think this bot would be able to accomplish and come up with an hourly rate for said jobs.

For my example, I'll use maid, averaging around $15 an hour.

If a maid works full time (40 hours per week) for a year, they will bring home $31,200 in the first year, and double that in the second year ($62,400).

As you can see, in our use case, the upper-end cost of a Bot at $50k is already lower than the 2-year cost of hiring a maid. But here's the deal - the bot is designed to work 16-hour shifts, which means 80-hour weeks. This is because bots don't need to rest, sleep, and go on strike. They're not human.

That means that in year one, the Bot will be able to do double the work of a maid, while costing 20% less.

And here's where it gets really nutty - starting in year two, the cost of the Bot is $0.

That's because you've already paid $50k up front. Instead of paying a maid $32k every year, for let's say 5 years, which equals to $160k over 5 years - you pay $50k up front for the bot, and you get DOUBLE THE WORK in those 5 years because it can work 16-hour shifts.

After 5 years, you would've paid the equivalent of $320k for a maid to do the same amount of work a Tesla Bot did for $50k.

That's a reduction of 85% in labor costs over 5 years.

There may be monthly subscription fees that Tesla could charge in order to do certain tasks - but these will likely pale in comparison to how much one would spend on the human-equivalent of each task.

If Tesla is able to manufacture millions of these per year, and they're capable of doing most physical human tasks, why would anyone pay a human to do said tasks?

This does two things - it dramatically lowers the cost of labor to a point where products and services could be a fraction of the cost vs where they are now, and it gives businesses no reason to hire a human for 7x the costs over 5 years to do the same exact job. The savings get larger the longer the Bot can work for without replacement.

Oh and by the way, the employer won't need to pay the bot any health insurance, 401k, and it will never be concerned about any HR risks.

The reality is that this future is much closer than many realize. It appears that Tesla has solved for the most difficult problem in human labor - an AI-brain that will automatically generate actions based on video that it analyzes in the real world. And with its manufacturing expertise, they should be able to crank out millions of these per year over the course of the next couple decades that should bring massive abundance but could also create massive chaos.

The robots are coming.

As a society, we need to begin thinking about the future that we want with a technology like this existing in our daily lives. More importantly, we need to put in safeguards so that this technology doesn't cause more harm than it does good.

We want a world of plenty for everyone, not a world of plenty for the few. Executed correctly, this technology can enable a future that few of us thought possible.

476

Barry Crisp @MrBarryCrisp

almost 3 years ago

@CalneTC @ryanwebb 👀

Who to follow

Ex-Oxford University IT admin, trained geologist & 'Net user since the 1980s.

Barry Crisp @MrBarryCrisp

almost 3 years ago

@BarackObama No @OfficialYJM then sadly.

Barry Crisp @MrBarryCrisp

over 3 years ago

@Alex__Briggs Literally did the same thing in my car this evening. Getting the spreader pin out was a bugger.

MrBarryCrisp retweeted

Wonder of Science

@wonderofscience

over 3 years ago

Contemplate the incredible vastness of space with this visualization of the size of our Milky Way galaxy. Credit: NASA/JPL-Caltech

102

579

MrBarryCrisp retweeted

Tom Wilson @thomas_m_wilson

over 3 years ago

SCOOP: Net energy gain in a fusion reaction has been a holy grail in science for decades. Now I’m told US scientists have done it. A massive breakthrough with revolutionary potential for clean power. US Energy Secretary to hold a press conference Tuesday https://t.co/aiyyRU3ehO

500

15K

844

Barry Crisp @MrBarryCrisp

over 3 years ago

My conversations with ChatGBT were predictably dull and machine like. It got the current (live) bassist of Whitesnake wrong.

Barry Crisp @MrBarryCrisp

over 3 years ago

@lucystonenow @ryanwebb The value to me is in getting realtime updates from respected engineers, team members, reporters etc during motorsport races. Some great insights that aren't available elsewhere currently.

Barry Crisp @MrBarryCrisp

over 4 years ago

@ryanwebb I must be getting harder words 😅

MrBarryCrisp retweeted

Colin McCarthy

@US_Stormwatch

over 4 years ago

Tonga's Hunga Tonga volcano just had one of the most violent volcano eruptions ever captured on satellite.

108K

38K

Barry Crisp @MrBarryCrisp

over 4 years ago

@suziperry @BBC_TopGear @GoodwoodRRC @ValeYellow46 Really great article @suziperry I felt myself welling up in joy as I read it such was the imagery of your words. Thanks for sharing.