3 ms·

Advanced auto-complete and sort-of right concept art is not going to disrupt the world nor the software industry. We have had algorithms trained on machine lear

Advanced auto-complete and sort-of right concept art is not going to disrupt the world nor the software industry. We have had algorithms trained on machine learning do IMO way more socially affecting things like decide what news and videos we watch, who gets approved for loans, what politicians say - but this gets way less press or thought. You will never replace software engineers just as you will never replace writers and artists no matter how “good” the machine generated stuff gets. There is no soul there and it only appears interesting at a superficial, “that’s sort of neat” level that doesn’t hold up. No one is buying master works of AI, they spend their weekend reading real books and blogs from real people. It’s 2022 - where is my self driving car? Where is the tens of millions of job losses from AI predicted that necessitated UBI (see the entire presidential thesis of Andrew Yang)? It does get a lot of hype and raises a lot of VC money though!

nmca 4y ago
https://bounded-regret.ghost.io/ai-forecasting-one-year-in/ https://bounded-regret.ghost.io/ai-forecasting-one-year-in/
kristiandupont 4y ago
May I ask if you have tried using Copilot? Because this sounds exactly like the kind of response I would have had half a year ago. It makes many mistakes and can by no means write code by itself. But that doesn't change the fact that it's orders of magnitude better at it than I thought possible.
- copenja 4y ago
  I'm not the parent post and I have not tried autopilot, but I do have a question. Does autopilot make you orders of magnitude more productive?
  - kristiandupont 4y ago
    No, it does not.
- AlotOfReading 4y ago
  One issue with copilot is that debugging broken code is far harder than writing it in the first place. We've seen similar transitions before though. When good optimizing compilers started appearing you had plenty of programmers like Mel [1] that said 'there's no art in it, a compiler couldn't do the kind of optimizations I do'. And yeah, it's true, but compilers got good enough that 99% of people stopped caring and hardcore optimization became a niche skill few people have. What does the industry look like when all that most programmers know is debugging generated code and what kinds of tools need to exist to support that? [1] http://catb.org/jargon/html/story-of-mel.html http://catb.org/jargon/html/story-of-mel.html

nmca 4y ago

https://bounded-regret.ghost.io/ai-forecasting-one-year-in/ https://bounded-regret.ghost.io/ai-forecasting-one-year-in/

kristiandupont 4y ago

May I ask if you have tried using Copilot? Because this sounds exactly like the kind of response I would have had half a year ago. It makes many mistakes and can by no means write code by itself. But that doesn't change the fact that it's orders of magnitude better at it than I thought possible.

copenja 4y ago
I'm not the parent post and I have not tried autopilot, but I do have a question. Does autopilot make you orders of magnitude more productive?
- kristiandupont 4y ago
  No, it does not.
AlotOfReading 4y ago
One issue with copilot is that debugging broken code is far harder than writing it in the first place. We've seen similar transitions before though. When good optimizing compilers started appearing you had plenty of programmers like Mel [1] that said 'there's no art in it, a compiler couldn't do the kind of optimizations I do'. And yeah, it's true, but compilers got good enough that 99% of people stopped caring and hardcore optimization became a niche skill few people have. What does the industry look like when all that most programmers know is debugging generated code and what kinds of tools need to exist to support that? [1] http://catb.org/jargon/html/story-of-mel.html http://catb.org/jargon/html/story-of-mel.html