6 ms·

> Did humans with computer replace humans without computer? Certainly for a whole host of tasks. > Did programmers with google search and stack overflow repla

> Did humans with computer replace humans without computer? Certainly for a whole host of tasks. > Did programmers with google search and stack overflow replace programmers without these? Also true for many tasks too, a programmer who refuses to use these resources for certain types of work will be an order of magnitude slower in many cases and will be replaced by programmers who aren't. LLMs are just beginning to become useful, but there will be a point where they're indispensable, those who refuse to use them will eventually find themselves falling behind. It's will eventually become equivalent to a programmer who refuses to move off punch cards or refuses to use modern version control.

pydry 3y ago
We're peak hype cycle for LLMs right now so we have an order of magnitude more ridiculous shit said about them than would otherwise be normal. The trough of disillusionment has yet to set in. They will be useful for a set of tasks but all of the people hyperventilating over this being the new industrial revolution might consider calming down a little. The high priests of capitalism have been blaming lost jobs on automation since forever - largely because this narrative makes a convenient, impersonal scapegoat.
- theLiminator 3y ago
  I'm curious what the "hype cycle" looked like for the discover of electricity, the invention of the combustion engine, or for the invention of the transistor. Who knows if the LLM paradigm will be more like those moments vs the failed hype of something like the hydrogen fuel cell.
  - pydry 3y ago
    I'm old enough to remember how it worked for the Internet. It was treated as a curiosity for nerds for a long while. The whole time I thought it was completely underrated and would change everything - thats why I decided to get into tech. Then there was a short, sharp "it's going to change everything" hype cycle that lasted no more than a few years followed by a trough of disillusionment that coincided with the recession that was around 2003 or so. I thought almost all parts of the cycle were exaggerated from the part where it was a curiosity for nerds to the "pets.com craze" where "the high street would disappear completely". With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications.
    - ivy137 3y ago
      a parlor trick lol. Here we are seeing statistics applied on abstract terms as human knowledge, how this is something with few practical applications?
      - FridgeSeal 3y ago
        Because the results and actual efficiency gains are flaky, inconsistent and mostly limited to boilerplate? GPT and friends are very impressive, yes; if you’re also impressed by your intern blindly copying stack overflow.
        ivy137 3y ago
        Maybe I am a fun of fuzzy logic, but I find gpt less flaky than most people that I read about, I think it will evolve. I feel gpt will even allow us to create new branch of mathematics applied at psychology. I can't understand why you compare it with an intern copy paste, it's simply absurd to compare. Looks just a way of reducing the argument and drawing parallelism where there are none. Yes, some people use it to program, No, not everyone uses it for programming
    - mixmastamyk 3y ago
      Then sadly big business and government figured out how to turn that promise into the dystopia of today.
    - theLiminator 3y ago
      > With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications. What makes you think this, I'm assuming you've tried gpt-4? I've personally gotten a lot of productivity out of it. Assuming it gets a lot better, I don't see how it's not going to minimally become an indispensable tool, and at maximum completely transform the future of humanity.
      - pydry 3y ago
  - _hypx 3y ago
    That’s a ridiculous statement. Hydrogen fuel cells will eventually power all cars.
- xcv123 3y ago
  No this is just getting started. https://lilianweng.github.io/posts/2023-06-23-agent/ https://lilianweng.github.io/posts/2023-06-23-agent/
softsound 3y ago
As someone who somewhat works with AI it's good if your data is good and terrible if it's not, my job is literally to clean up AI failures, and due to bad data or dealing with interpretation a lot can be done with data good or bad. My work helps retrain the model, but sometimes because of bad management we have unclear answers on how to interpret the data and this leads to some coworkers training it wrong and you'll see this linger between projects. Now if the people were better trained the data would be too but oh well. Someone wants to cut costs. I think this kind of issue will always exist no matter the model, you can't really make up data you don't have (I mean you certainly can but it all has to be figured out how to fix it when it does just start guessing poorly) and sometimes there are no answers but bad answers in some edge cases.
wruza 3y ago
Text editing and version control solved very specific problems. E.g. the former allows quick iteration and the latter allows to analyze changes and work in parallel. Which problems do programmers have that only AI could help with?

pydry 3y ago

We're peak hype cycle for LLMs right now so we have an order of magnitude more ridiculous shit said about them than would otherwise be normal. The trough of disillusionment has yet to set in. They will be useful for a set of tasks but all of the people hyperventilating over this being the new industrial revolution might consider calming down a little. The high priests of capitalism have been blaming lost jobs on automation since forever - largely because this narrative makes a convenient, impersonal scapegoat.

theLiminator 3y ago
I'm curious what the "hype cycle" looked like for the discover of electricity, the invention of the combustion engine, or for the invention of the transistor. Who knows if the LLM paradigm will be more like those moments vs the failed hype of something like the hydrogen fuel cell.
- pydry 3y ago
  I'm old enough to remember how it worked for the Internet. It was treated as a curiosity for nerds for a long while. The whole time I thought it was completely underrated and would change everything - thats why I decided to get into tech. Then there was a short, sharp "it's going to change everything" hype cycle that lasted no more than a few years followed by a trough of disillusionment that coincided with the recession that was around 2003 or so. I thought almost all parts of the cycle were exaggerated from the part where it was a curiosity for nerds to the "pets.com craze" where "the high street would disappear completely". With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications.
  - ivy137 3y ago
    a parlor trick lol. Here we are seeing statistics applied on abstract terms as human knowledge, how this is something with few practical applications?
    - FridgeSeal 3y ago
      Because the results and actual efficiency gains are flaky, inconsistent and mostly limited to boilerplate? GPT and friends are very impressive, yes; if you’re also impressed by your intern blindly copying stack overflow.
      - ivy137 3y ago
        Maybe I am a fun of fuzzy logic, but I find gpt less flaky than most people that I read about, I think it will evolve. I feel gpt will even allow us to create new branch of mathematics applied at psychology. I can't understand why you compare it with an intern copy paste, it's simply absurd to compare. Looks just a way of reducing the argument and drawing parallelism where there are none. Yes, some people use it to program, No, not everyone uses it for programming
  - mixmastamyk 3y ago
    Then sadly big business and government figured out how to turn that promise into the dystopia of today.
  - theLiminator 3y ago
    > With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications. What makes you think this, I'm assuming you've tried gpt-4? I've personally gotten a lot of productivity out of it. Assuming it gets a lot better, I don't see how it's not going to minimally become an indispensable tool, and at maximum completely transform the future of humanity.
    - pydry 3y ago
- _hypx 3y ago
  That’s a ridiculous statement. Hydrogen fuel cells will eventually power all cars.
xcv123 3y ago
No this is just getting started. https://lilianweng.github.io/posts/2023-06-23-agent/ https://lilianweng.github.io/posts/2023-06-23-agent/

softsound 3y ago

As someone who somewhat works with AI it's good if your data is good and terrible if it's not, my job is literally to clean up AI failures, and due to bad data or dealing with interpretation a lot can be done with data good or bad. My work helps retrain the model, but sometimes because of bad management we have unclear answers on how to interpret the data and this leads to some coworkers training it wrong and you'll see this linger between projects. Now if the people were better trained the data would be too but oh well. Someone wants to cut costs. I think this kind of issue will always exist no matter the model, you can't really make up data you don't have (I mean you certainly can but it all has to be figured out how to fix it when it does just start guessing poorly) and sometimes there are no answers but bad answers in some edge cases.

wruza 3y ago

Text editing and version control solved very specific problems. E.g. the former allows quick iteration and the latter allows to analyze changes and work in parallel. Which problems do programmers have that only AI could help with?