4 ms·

I don't think it's going to be as bad as everyone is saying in this comment thread. Current AI's not accurate enough to replace most human work and only showing

I don't think it's going to be as bad as everyone is saying in this comment thread. Current AI's not accurate enough to replace most human work and only showing incremental improvements lately. A new technological strategy is probably needed to achieve human-level capability. The jobs at risk / already dead from AI are things like internet ghost writing / SEO spam writing and customer service roles for cheap companies. For the rest of us it's a (very useful) tool, not a competitor.

somenameforme 3y ago
I'd concur and also add one other aspect here. The reason a person hires a e.g. janitor is not because they themselves are incapable of janitorial work. It's because they have different tasks that they would prefer to dedicate their energy to. Even if there were absolutely perfect LLMs capable of e.g. generating code precisely to specification, you still end up with the issue of actually creating said specification, deploying the result, managing with various local constraints/limitations, updating all of this when it turns out there's an unforeseen issue, and so on. Even if its work that a wider range of people could do, it's still work that's going to need to be done, and it's not going to be done by a chatbot. So it seems that even for many jobs that LLMs could ostensibly replace, it's more likely to widen the labor pool rather than literally replace people. If anything it might increase the number of businesses and opportunities, because it would enable the breadth of domains that micro and small companies could perform in.
- analognoise 3y ago
  It will widen the labor pool until it suddenly doesn’t anymore, right? “Horses Need Not Apply” was all about that - the point being new devices were introduced until suddenly horses became pets rather than working animals (on a farm), and our insistence that technological change creates new jobs (“creative destruction”) wasn’t well founded.
  - somenameforme 2y ago
    I'm not arguing that more tech advances create more jobs. I'm arguing that what we're trying to automate would be insufficient to obsolete many of the jobs people fear it could, even if successful. And I think that's a very big if. Vehicles were quite unique in that they were capable of entirely replacing their predecessor for simple transportation tasks. But in e.g. software, the programming tasks LLMs may one day be able to carry out are but one aspect of the job. Getting back to janitors for instance, you're not going to replace a janitor with a roomba.
jmholla 3y ago
> Current AI's not accurate enough to replace most human work and only showing incremental improvements lately. But I don't think that's the real risk. The real risk is people in charge of jobs that believe AI can replace their employees, regardless of its true capabilities.
- pfdietz 3y ago
  Like any of the myriad of delusions that have swept through market economies, that's a self-limiting risk.
  - jmholla 3y ago
    Normally, I'd agree. But with the centralization and corresponding dominance of so many services, I worry many companies can make that choice and not have market effects correct them.
bearjaws 3y ago
I am in the same boat as you. LLMs cannot even solve math problems, and we expect them to do tasks far more complex e.g. software engineering. Attempts to make them good at math have shown its not easy, and often times sacrifices language quality. How do we expect something that cannot solve y=mx+b reliably for any input >100 to write more than a todo list app? A lot of companies/investors are just assuming its a training data problem, but we can already produce massive amounts of robust math training data, and LLMs still aren't good at math. At least this is my understanding, I was doing some research into Kahnmigo and its math tutoring capabilities and went down a rabbit hole about LLM math.
- randomdata 3y ago
  > and we expect them to do tasks far more complex e.g. software engineering. Who expects that? Even if LLMs produced perfect output every time, they are still just a fancy compiler. You continue to need a software engineer to write the code. Some expect that they will enable software to go the way of the elevator, which is to say that modern elevators didn't eliminate the elevator operator – they made everyone the elevator operator. This is quite possible, and has arguably already happened at least in some niche cases. But that is an expansion of the pool of human software engineers, not an LLM doing software engineering. Maybe there is some kind of technology out there that will someday take on the tasks of software engineering, but not LLMs. They cannot do software engineering any more than a C++ compiler can do software engineering.
  - bearjaws 3y ago
    > Who expects that? https://www.cognition-labs.com/introducing-devin https://www.cognition-labs.com/introducing-devin SA: “Coding is probably the single area from a productivity gain we’re most excited about today" Nvidia CEO: "Don't learn to code" The whole group of companies working on it are all pushing that it will make programming redundant, but they don't even make a Ti-83 redundant.
    - randomdata 3y ago
      Not a whole lot of substance on that page, but the video shows that you first have to write a program before it will do anything. It too just looks like another compiler (and execution environment). An advanced compiler, to be sure, but there is no sign of it doing the job of the software engineer.

somenameforme 3y ago

I'd concur and also add one other aspect here. The reason a person hires a e.g. janitor is not because they themselves are incapable of janitorial work. It's because they have different tasks that they would prefer to dedicate their energy to. Even if there were absolutely perfect LLMs capable of e.g. generating code precisely to specification, you still end up with the issue of actually creating said specification, deploying the result, managing with various local constraints/limitations, updating all of this when it turns out there's an unforeseen issue, and so on. Even if its work that a wider range of people could do, it's still work that's going to need to be done, and it's not going to be done by a chatbot. So it seems that even for many jobs that LLMs could ostensibly replace, it's more likely to widen the labor pool rather than literally replace people. If anything it might increase the number of businesses and opportunities, because it would enable the breadth of domains that micro and small companies could perform in.

analognoise 3y ago
It will widen the labor pool until it suddenly doesn’t anymore, right? “Horses Need Not Apply” was all about that - the point being new devices were introduced until suddenly horses became pets rather than working animals (on a farm), and our insistence that technological change creates new jobs (“creative destruction”) wasn’t well founded.
- somenameforme 2y ago
  I'm not arguing that more tech advances create more jobs. I'm arguing that what we're trying to automate would be insufficient to obsolete many of the jobs people fear it could, even if successful. And I think that's a very big if. Vehicles were quite unique in that they were capable of entirely replacing their predecessor for simple transportation tasks. But in e.g. software, the programming tasks LLMs may one day be able to carry out are but one aspect of the job. Getting back to janitors for instance, you're not going to replace a janitor with a roomba.

jmholla 3y ago

> Current AI's not accurate enough to replace most human work and only showing incremental improvements lately. But I don't think that's the real risk. The real risk is people in charge of jobs that believe AI can replace their employees, regardless of its true capabilities.

pfdietz 3y ago
Like any of the myriad of delusions that have swept through market economies, that's a self-limiting risk.
- jmholla 3y ago
  Normally, I'd agree. But with the centralization and corresponding dominance of so many services, I worry many companies can make that choice and not have market effects correct them.

bearjaws 3y ago

I am in the same boat as you. LLMs cannot even solve math problems, and we expect them to do tasks far more complex e.g. software engineering. Attempts to make them good at math have shown its not easy, and often times sacrifices language quality. How do we expect something that cannot solve y=mx+b reliably for any input >100 to write more than a todo list app? A lot of companies/investors are just assuming its a training data problem, but we can already produce massive amounts of robust math training data, and LLMs still aren't good at math. At least this is my understanding, I was doing some research into Kahnmigo and its math tutoring capabilities and went down a rabbit hole about LLM math.

randomdata 3y ago
> and we expect them to do tasks far more complex e.g. software engineering. Who expects that? Even if LLMs produced perfect output every time, they are still just a fancy compiler. You continue to need a software engineer to write the code. Some expect that they will enable software to go the way of the elevator, which is to say that modern elevators didn't eliminate the elevator operator – they made everyone the elevator operator. This is quite possible, and has arguably already happened at least in some niche cases. But that is an expansion of the pool of human software engineers, not an LLM doing software engineering. Maybe there is some kind of technology out there that will someday take on the tasks of software engineering, but not LLMs. They cannot do software engineering any more than a C++ compiler can do software engineering.
- bearjaws 3y ago
  > Who expects that? https://www.cognition-labs.com/introducing-devin https://www.cognition-labs.com/introducing-devin SA: “Coding is probably the single area from a productivity gain we’re most excited about today" Nvidia CEO: "Don't learn to code" The whole group of companies working on it are all pushing that it will make programming redundant, but they don't even make a Ti-83 redundant.
  - randomdata 3y ago
    Not a whole lot of substance on that page, but the video shows that you first have to write a program before it will do anything. It too just looks like another compiler (and execution environment). An advanced compiler, to be sure, but there is no sign of it doing the job of the software engineer.