5 ms·

I don't see the current software industry as sustainable. First off, I don't think it solves as many problems as the salesmen think it does, it just offloads co

I don't see the current software industry as sustainable. First off, I don't think it solves as many problems as the salesmen think it does, it just offloads complexity onto someone else, and more importantly, there's no stability. You can't just build something and it works long term with minimal maintenance. There is probably no software that exists today that will be able to continue running 100 years from now. I think we are in for a big wall we hit where software companies go bankrupt, people move away from it for lots of uses, followed by develpment of more stable tooling, after which people only use it where they have to, such as automating organization of complex systems.

valine 3y ago
Or, more likely, the march towards better abstraction ends with a natural language interface where problem descriptions can be effortlessly translated into functional code. We’re already in spitting distance of this with current language models. I don’t think you even need AGI to make this happen, a large language/image model gets you most of the way there.
- friend_and_foe 3y ago
  Well, that falls under the "better tooling" I referenced. We have to have some set of tools as yet unimagined that are so robust that they don't need constant changes. I have to be able to use a piece of software and be able to explain to my grandkids how to use it when I'm old. A part of the fast changing problem is that we are still developing the technology, but that means once we can breathe and things become more stable and change slower we will look back on a lot of mistakes and pain. There are a lot of people invested in the current unsustainable state of affairs and this can't go on forever, those people are going to lose their shirts.
- WesolyKubeczek 3y ago
  Right now humans are quite shit at translation of natural panguage into code. Other humans are shit at stating their requirements in any language at all. Don’t keep your hopes high.
  - valine 3y ago
    GPT-4 can write html from a messy napkin sketch. These models won’t need formal requirements. “I’m having a problem with A, B and C” will likely be all the model needs from you. Your argument was very common and quite persuasive two years ago. Not so much today.
  - LoganDark 3y ago
    https://www.commitstrip.com/en/2016/08/25/a-very-comprehensive-and-precise-spec https://www.commitstrip.com/en/2016/08/25/a-very-comprehensi...
    - valine 3y ago
      This 2016 comic failed to predict GPTs. It turns out the same model that writes your code can also write the specification. I don’t blame the author, it’s nearly impossible to predict future technology. Pretending that GPTs don’t exist on the other hand, now that’s just silly.
      - LoganDark 3y ago
        I think what's more likely to happen is that people will be able to turn a conversation into code instead of just a problem statement. In fact, ChatGPT already does that, although you still need to have some skill in order to properly integrate and test the code that it gives you.
        valine 3y ago
        Sure, the more information you give the model the better. While our current models generate imperfect code, GPT4 code is much more usable out of the box than GPT3.5 code. If that trend continues we will absolutely get to the point where mistakes are the exception not the rule.
        LoganDark 3y ago
        I wonder if once (better) multimodal models are a thing, a model might actually be able to close the feedback loop itself. It won't magically generate a working product the first time but it might be able to test it and iterate on it automatically, depending on what you're trying to do. You then just have to make sure it's iterating on the thing you actually want.
- cubefox 3y ago
  Yeah. It seems Hanson hasn't considered the effects of AI. It can plausibly compensate the productivity loss from a shrinking population.
- danaris 3y ago
  It's hard enough for humans to take natural-language descriptions of a problem to be solved (that is well within human capabilities) and turn them into human-workable solutions without a significant process of back and forth to get the requirements clear. Getting a computer, within the next decade or two, to do better, doesn't seem realistic—not because I think it is impossible that we will have a computer fully capable of parsing natural human language into internal representations of the concepts it represents with high fidelity, but simply because humans are bad at communicating. Tell me you've never, ever had a project described to you by someone who was horribly nonspecific—or even outright wrong—in their description of what they need. (If you can truthfully do so, I can easily find you a dozen people who can't—starting with myself.) Human language is woefully inadequate to convey the breadth and depth of human experience, needs, wants, requirements, etc. This is why we invented programming languages in the first place.
RetroTechie 3y ago
The end of Moore's law is a blessing in disguise imho. For too long, software has been written to make inefficient use of hardware, and/or designed with only next quarter's profit in mind. Based on the implicit assumption that faster hardware is just around the corner, and will make crappy software look decent. Security, frugal use of resources, reliability, maintenance, most anything was sacrificed in order to put products/services out there asap. Once out there, a continuous upgrade treadmill, and 3..5y later it's ewaste / bitrotted away. We can put more compute in a few mm^2 than any human ever needs, at ridiculous low cost. It's about time the software world gets a breather, makes slow UIs fast again, and does some rework to dump the GBs of crap that populates today's systems.

valine 3y ago

Or, more likely, the march towards better abstraction ends with a natural language interface where problem descriptions can be effortlessly translated into functional code. We’re already in spitting distance of this with current language models. I don’t think you even need AGI to make this happen, a large language/image model gets you most of the way there.

friend_and_foe 3y ago
Well, that falls under the "better tooling" I referenced. We have to have some set of tools as yet unimagined that are so robust that they don't need constant changes. I have to be able to use a piece of software and be able to explain to my grandkids how to use it when I'm old. A part of the fast changing problem is that we are still developing the technology, but that means once we can breathe and things become more stable and change slower we will look back on a lot of mistakes and pain. There are a lot of people invested in the current unsustainable state of affairs and this can't go on forever, those people are going to lose their shirts.
WesolyKubeczek 3y ago
Right now humans are quite shit at translation of natural panguage into code. Other humans are shit at stating their requirements in any language at all. Don’t keep your hopes high.
- valine 3y ago
  GPT-4 can write html from a messy napkin sketch. These models won’t need formal requirements. “I’m having a problem with A, B and C” will likely be all the model needs from you. Your argument was very common and quite persuasive two years ago. Not so much today.
- LoganDark 3y ago
  https://www.commitstrip.com/en/2016/08/25/a-very-comprehensive-and-precise-spec https://www.commitstrip.com/en/2016/08/25/a-very-comprehensi...
  - valine 3y ago
    This 2016 comic failed to predict GPTs. It turns out the same model that writes your code can also write the specification. I don’t blame the author, it’s nearly impossible to predict future technology. Pretending that GPTs don’t exist on the other hand, now that’s just silly.
    - LoganDark 3y ago
      I think what's more likely to happen is that people will be able to turn a conversation into code instead of just a problem statement. In fact, ChatGPT already does that, although you still need to have some skill in order to properly integrate and test the code that it gives you.
      - valine 3y ago
        Sure, the more information you give the model the better. While our current models generate imperfect code, GPT4 code is much more usable out of the box than GPT3.5 code. If that trend continues we will absolutely get to the point where mistakes are the exception not the rule.
        LoganDark 3y ago
        I wonder if once (better) multimodal models are a thing, a model might actually be able to close the feedback loop itself. It won't magically generate a working product the first time but it might be able to test it and iterate on it automatically, depending on what you're trying to do. You then just have to make sure it's iterating on the thing you actually want.
cubefox 3y ago
Yeah. It seems Hanson hasn't considered the effects of AI. It can plausibly compensate the productivity loss from a shrinking population.
danaris 3y ago
It's hard enough for humans to take natural-language descriptions of a problem to be solved (that is well within human capabilities) and turn them into human-workable solutions without a significant process of back and forth to get the requirements clear. Getting a computer, within the next decade or two, to do better, doesn't seem realistic—not because I think it is impossible that we will have a computer fully capable of parsing natural human language into internal representations of the concepts it represents with high fidelity, but simply because humans are bad at communicating. Tell me you've never, ever had a project described to you by someone who was horribly nonspecific—or even outright wrong—in their description of what they need. (If you can truthfully do so, I can easily find you a dozen people who can't—starting with myself.) Human language is woefully inadequate to convey the breadth and depth of human experience, needs, wants, requirements, etc. This is why we invented programming languages in the first place.

RetroTechie 3y ago

The end of Moore's law is a blessing in disguise imho. For too long, software has been written to make inefficient use of hardware, and/or designed with only next quarter's profit in mind. Based on the implicit assumption that faster hardware is just around the corner, and will make crappy software look decent. Security, frugal use of resources, reliability, maintenance, most anything was sacrificed in order to put products/services out there asap. Once out there, a continuous upgrade treadmill, and 3..5y later it's ewaste / bitrotted away. We can put more compute in a few mm^2 than any human ever needs, at ridiculous low cost. It's about time the software world gets a breather, makes slow UIs fast again, and does some rework to dump the GBs of crap that populates today's systems.