4 ms·

> outperforms humans at most economically valuable work "most" but not all economically valuable work. ChatGPT (or what follows) can't swing a hammer and it c

> outperforms humans at most economically valuable work "most" but not all economically valuable work. ChatGPT (or what follows) can't swing a hammer and it can't apply spackle. Maybe I should start looking into the trades.

lucasyvas 3y ago
It's not off the cards. A lot of what we produce is frankly useless from a core value perspective compared to the outputs of physical labor. If it doesn't produce or dramatically aid in the production of Food, Water, Shelter, Energy, or Transport... it will never be as intrinsically valuable. The vast majority of work in those categories requires irreplaceable physical labor.
- jstummbillig 3y ago
  We need to interface with the world, but the minimum building blocks required to do that for all industries with just software are going to get smaller and smaller. When you have a device that can manipulate atoms as computers can manipulate bits, you will be able to do everything with software. We are probably not going to write it, but I assume as long as we are in control, we are going to drive the design. Software is most certainly the end game.
  - RandomLensman 3y ago
    Are we back to some of the magic requiring ideas of the nanotechnology hype?
- Kapura 3y ago
  Extremely weird, or maybe not weird at all, that you don't include art of any form in your list of intrinsically valuable things.
  - lucasyvas 3y ago
    Art is something you can survive without, but can't practically live without - I did omit it on purpose because expression seems hard to categorize - it's what we live for, but not how we live.
- sebastiennight 3y ago
  Let's look at the top list of most valuable companies in the world right now: 1. Apple 2. Microsoft 3. Saudi Aramco 4. Alphabet (Google) 5. Amazon 6. NVIDIA 7. Meta (Facebook) 8. Berkshire Hathaway 9. Tesla 10. Eli Lilly 11. TSMC 12. Visa How many would you reckon are dramatically involved in "Food, Water, Shelter, Energy, or Transport" in a way that software can't enhance, multiply or replace?
  - lucasyvas 3y ago
    The key words here are enhance and multiply (I don't agree with replace) - of those on your list, I'd say that Tesla and Eli Lilly stand to contribute the most tangible benefit to our lives if they continue to execute and improve. Saudi Aramco, while not working in a line of energy we want to pursue long term, also belongs in there for now. Apple, Microsoft, Alphabet, Amazon, NVIDIA, Meta, Berkshire Hathaway, TSMC, and Visa do not produce anywhere near the same kind of core, intrinsic value. We have a fundamental disconnect between monetary value and actual value to society. I'm suggesting that our capital has been invested extremely poorly for at least the last two decades - chasing easy software wins instead of building hardware we need to solve real problems and putting humans to work to help construct it. These companies build products and services that enable business, commerce, hardware and software that enhance our experience in the world as we know it (including those core categories), but the world as we know it could change in an instant. If everything were to go to shit, the most valuable are Food, Water, Shelter, Energy, or Transport (and as another poster pointed out, probably Art although it's kind of in a league of its own). They've always had the most intrinsic value. And while we need software to enhance a lot of these, the physical laborers are far less replaceable and therefore far more valuable. Technology is much farther away from replacing people in meatspace than cyberspace. Apple is not stupid - they are probably best positioned to dominate computing over the next decade at least. They are computing company that specializes in enhancing experience, but there is a looming energy crisis and they know it. That's why their CPUs are focusing so heavily on performance per watt. We are going to need radically more efficient hardware that is cheaper to produce to meaningfully deploy computers to solve out problems in the future. If datacenters can not become sourced entirely by renewables, the your pocket computer (and personal AI) is going to be more sustainable and therefore more valuable.
avereveard 3y ago
It's getting close at being able to program a robotic arm to swing an hammer, as soon as vision can do video it'll be able to debug it. All considered were somewhat close to embodiment. It will take quite a while to make it real time, but still. Edit: heck I believe that if you have a car moving at 1cm per 20 second that's sending a picture every 20 second to gpt4v and some glue code to send and receive from gpt4v it'll will be able to self drive.
- busyant 3y ago
  > t's getting close at being able to program a robotic arm to swing an hammer, as soon as vision can do video it'll be able to debug it Agreed. My original comment was simply that ChatGPT (or whatever) can't swing a hammer yet. The whole thing feels like an oncoming tsunami and I'm just trying to figure out how to scramble to higher ground (while it lasts).
thatfrenchguy 3y ago
I mean, folks who wrote that text likely have been removed from physical work for so many generations that they forgot how valuable it is.
namibj 3y ago
Don't underestimate a robot who can actually think. In many cases physical labor is about having closed-visual-feedback-loop (hand-eye-coordination) robots that are extremely sample-efficient to train up to useful skill, and still (but not necessarily extremely so) sample efficient to train to near their maximum performance at that task. They can even do multi-modal transfer learning: have deviations from a previously learned process explained in words it's diagrams, then start directly zero-shot-like at a performance level that won't waste ingredients, with all the normal learning speed aspects still remaining.

lucasyvas 3y ago

It's not off the cards. A lot of what we produce is frankly useless from a core value perspective compared to the outputs of physical labor. If it doesn't produce or dramatically aid in the production of Food, Water, Shelter, Energy, or Transport... it will never be as intrinsically valuable. The vast majority of work in those categories requires irreplaceable physical labor.

jstummbillig 3y ago
We need to interface with the world, but the minimum building blocks required to do that for all industries with just software are going to get smaller and smaller. When you have a device that can manipulate atoms as computers can manipulate bits, you will be able to do everything with software. We are probably not going to write it, but I assume as long as we are in control, we are going to drive the design. Software is most certainly the end game.
- RandomLensman 3y ago
  Are we back to some of the magic requiring ideas of the nanotechnology hype?
Kapura 3y ago
Extremely weird, or maybe not weird at all, that you don't include art of any form in your list of intrinsically valuable things.
- lucasyvas 3y ago
  Art is something you can survive without, but can't practically live without - I did omit it on purpose because expression seems hard to categorize - it's what we live for, but not how we live.
sebastiennight 3y ago
Let's look at the top list of most valuable companies in the world right now: 1. Apple 2. Microsoft 3. Saudi Aramco 4. Alphabet (Google) 5. Amazon 6. NVIDIA 7. Meta (Facebook) 8. Berkshire Hathaway 9. Tesla 10. Eli Lilly 11. TSMC 12. Visa How many would you reckon are dramatically involved in "Food, Water, Shelter, Energy, or Transport" in a way that software can't enhance, multiply or replace?
- lucasyvas 3y ago
  The key words here are enhance and multiply (I don't agree with replace) - of those on your list, I'd say that Tesla and Eli Lilly stand to contribute the most tangible benefit to our lives if they continue to execute and improve. Saudi Aramco, while not working in a line of energy we want to pursue long term, also belongs in there for now. Apple, Microsoft, Alphabet, Amazon, NVIDIA, Meta, Berkshire Hathaway, TSMC, and Visa do not produce anywhere near the same kind of core, intrinsic value. We have a fundamental disconnect between monetary value and actual value to society. I'm suggesting that our capital has been invested extremely poorly for at least the last two decades - chasing easy software wins instead of building hardware we need to solve real problems and putting humans to work to help construct it. These companies build products and services that enable business, commerce, hardware and software that enhance our experience in the world as we know it (including those core categories), but the world as we know it could change in an instant. If everything were to go to shit, the most valuable are Food, Water, Shelter, Energy, or Transport (and as another poster pointed out, probably Art although it's kind of in a league of its own). They've always had the most intrinsic value. And while we need software to enhance a lot of these, the physical laborers are far less replaceable and therefore far more valuable. Technology is much farther away from replacing people in meatspace than cyberspace. Apple is not stupid - they are probably best positioned to dominate computing over the next decade at least. They are computing company that specializes in enhancing experience, but there is a looming energy crisis and they know it. That's why their CPUs are focusing so heavily on performance per watt. We are going to need radically more efficient hardware that is cheaper to produce to meaningfully deploy computers to solve out problems in the future. If datacenters can not become sourced entirely by renewables, the your pocket computer (and personal AI) is going to be more sustainable and therefore more valuable.

avereveard 3y ago

It's getting close at being able to program a robotic arm to swing an hammer, as soon as vision can do video it'll be able to debug it. All considered were somewhat close to embodiment. It will take quite a while to make it real time, but still. Edit: heck I believe that if you have a car moving at 1cm per 20 second that's sending a picture every 20 second to gpt4v and some glue code to send and receive from gpt4v it'll will be able to self drive.

busyant 3y ago
> t's getting close at being able to program a robotic arm to swing an hammer, as soon as vision can do video it'll be able to debug it Agreed. My original comment was simply that ChatGPT (or whatever) can't swing a hammer yet. The whole thing feels like an oncoming tsunami and I'm just trying to figure out how to scramble to higher ground (while it lasts).

thatfrenchguy 3y ago

I mean, folks who wrote that text likely have been removed from physical work for so many generations that they forgot how valuable it is.

namibj 3y ago

Don't underestimate a robot who can actually think. In many cases physical labor is about having closed-visual-feedback-loop (hand-eye-coordination) robots that are extremely sample-efficient to train up to useful skill, and still (but not necessarily extremely so) sample efficient to train to near their maximum performance at that task. They can even do multi-modal transfer learning: have deviations from a previously learned process explained in words it's diagrams, then start directly zero-shot-like at a performance level that won't waste ingredients, with all the normal learning speed aspects still remaining.