3 ms·

My view is that even if AI can program really well, it can't architect full solutions, you still need to tell it specifics and test/debug its output (at least f

by _8j50 4y ago

My view is that even if AI can program really well, it can't architect full solutions, you still need to tell it specifics and test/debug its output (at least for now). It's like if you're a farmer and you got tractors, you don't need to till and harvest anymore but you still gotta drive it and press buttons and even if tractors drove themselves and pressed their own buttons you gotta supervise them and maintain them. I think people being less eager to be programmers will only help with increasing wages lol

famahar 4y ago
A single crop harvester can replace 30 farm workers. If programming is going the same route as your farming example then we're in for a shock. The complexity and scale of software increases every year so I wonder if AI will be able to keep up or if there will always be a need for specialized roles.
- realusername 4y ago
  Coding is a small percentage of a programming job, I don't write more than 100 lines / day, GPT tools are also very late to the party of code generation, it's been 40 years of development in this area.
- _8j50 4y ago
  Yeah but not everyone affords a crop harvester and even then you have less farm workers not no farm workers. I think there will be less demand for programmers and more demand at lesser pay for the top quintile of talented coders. There will also be ML managing/engineering/monitoring type jobs that will require you to know how to code. Various other disciplines will also have knowing python and shell scripting a requirement. The only thing that will get decimated is the need for a 10:1 ratio (making that up for the record) between coders that stitch software and suck at debugging and coders that sre debugging ninjas that can identify issues GPT* hallucinates. The next stage of decimation is a long way off when GPT is able to be integrated into the production monitoring tooling, debug on it's own and answer questions like "why is your code slow under these circumstances? And why does it randomly crash?" Right now operations/sre/users ask that of the programmer. I don't believe GPT is anywhere near being able to reliably answer those questions because if it were, why emit such code to begin with? You would be able to prompt it to generate perfect code if it could do that, the code I have seen is impressive but full of flaws even a collegr freshman wouldn't make. To me the question is the same as dalle-e replacing artists. Even whe the drawing is not weird or just messed up it still does not "wow" you, i mean technically it qualifies and for a machine it is impressive but it just feels mediocre to the most part. I think in reality what you will see is GPT assisted IDEs giving new meaning to "code completion".
whateveracct 4y ago
And you still need to run the software. ChatGPT isn't infallible and you will still push bugs or hit sudden scaling or perf issues. Maybe it can help debug at 2am but honestly I'd rather be debugging a system I built rather than one I partially copy-pasted in and didn't fully understand. The old fashioned way has the benefit of growing your own mind and making you "stronger and smarter"

by _8j50 4y ago

famahar 4y ago

A single crop harvester can replace 30 farm workers. If programming is going the same route as your farming example then we're in for a shock. The complexity and scale of software increases every year so I wonder if AI will be able to keep up or if there will always be a need for specialized roles.

realusername 4y ago
Coding is a small percentage of a programming job, I don't write more than 100 lines / day, GPT tools are also very late to the party of code generation, it's been 40 years of development in this area.
_8j50 4y ago
Yeah but not everyone affords a crop harvester and even then you have less farm workers not no farm workers. I think there will be less demand for programmers and more demand at lesser pay for the top quintile of talented coders. There will also be ML managing/engineering/monitoring type jobs that will require you to know how to code. Various other disciplines will also have knowing python and shell scripting a requirement. The only thing that will get decimated is the need for a 10:1 ratio (making that up for the record) between coders that stitch software and suck at debugging and coders that sre debugging ninjas that can identify issues GPT* hallucinates. The next stage of decimation is a long way off when GPT is able to be integrated into the production monitoring tooling, debug on it's own and answer questions like "why is your code slow under these circumstances? And why does it randomly crash?" Right now operations/sre/users ask that of the programmer. I don't believe GPT is anywhere near being able to reliably answer those questions because if it were, why emit such code to begin with? You would be able to prompt it to generate perfect code if it could do that, the code I have seen is impressive but full of flaws even a collegr freshman wouldn't make. To me the question is the same as dalle-e replacing artists. Even whe the drawing is not weird or just messed up it still does not "wow" you, i mean technically it qualifies and for a machine it is impressive but it just feels mediocre to the most part. I think in reality what you will see is GPT assisted IDEs giving new meaning to "code completion".

whateveracct 4y ago

And you still need to run the software. ChatGPT isn't infallible and you will still push bugs or hit sudden scaling or perf issues. Maybe it can help debug at 2am but honestly I'd rather be debugging a system I built rather than one I partially copy-pasted in and didn't fully understand. The old fashioned way has the benefit of growing your own mind and making you "stronger and smarter"