User Power
Value/Post Ratio
125%
- Sep 11, 2018
- 657
- 821
I think this ChatGPT is impressive, but not the everyone will become unemployed fear-mongering some people are into.
I tried to use it to generate copy and I wasn't happy. Lifeless, generic, boring copy. Good starting point and source of ideas, so definitely useful, but still.
I can't see this replacing programmers any time soon either. Even if it helps with small pieces of code, or maybe converting code from some languages to others (although I read that it's got issues with some languages that are less mainstream or have less common paradigms, like Clojure - haven't tried myself though), the big picture of building a complex system is probably way above this AI technology and any technology in the near future, if possible at all. It might reduce the junior man time needed, but I don't see this replacing more experienced developers that make a difference. "Build me a website like Facebook that generates X billions in revenue" is not going to work; not now, not in decades I think. Not even a simpler web application (static websites are a different story though, but that's not "coding" anyway).
Understanding requirements is the most difficult part, and making sure whoever decides those requirements doesn't change them 10 times or misses 50% of them to start with. I'd like to see how this tech works when requirements change and the code needs to accommodate for this. I can see it generating fresh code accurately, probably replacing some methodologies like TDD, but not quite evolving existing code accurately. I've seen that it can help fixing bugs. Not sure about the complexity though, and I assume it's something like "program crashes, here's the offending line, I'll implement one of the most common fixes and see if it's fixed, otherwise try the next one". I bet languages such as Rust, that make it more difficult to shot yourself in the foot, won't benefit from ChatGPT much as most runtime errors are made impossible by the compiler. In any case, fixing bugs of whatever kind looks easier to achieve than adapting existing code for new functionality.
I might be wrong in 20 years, let's see.
I tried to use it to generate copy and I wasn't happy. Lifeless, generic, boring copy. Good starting point and source of ideas, so definitely useful, but still.
I can't see this replacing programmers any time soon either. Even if it helps with small pieces of code, or maybe converting code from some languages to others (although I read that it's got issues with some languages that are less mainstream or have less common paradigms, like Clojure - haven't tried myself though), the big picture of building a complex system is probably way above this AI technology and any technology in the near future, if possible at all. It might reduce the junior man time needed, but I don't see this replacing more experienced developers that make a difference. "Build me a website like Facebook that generates X billions in revenue" is not going to work; not now, not in decades I think. Not even a simpler web application (static websites are a different story though, but that's not "coding" anyway).
Understanding requirements is the most difficult part, and making sure whoever decides those requirements doesn't change them 10 times or misses 50% of them to start with. I'd like to see how this tech works when requirements change and the code needs to accommodate for this. I can see it generating fresh code accurately, probably replacing some methodologies like TDD, but not quite evolving existing code accurately. I've seen that it can help fixing bugs. Not sure about the complexity though, and I assume it's something like "program crashes, here's the offending line, I'll implement one of the most common fixes and see if it's fixed, otherwise try the next one". I bet languages such as Rust, that make it more difficult to shot yourself in the foot, won't benefit from ChatGPT much as most runtime errors are made impossible by the compiler. In any case, fixing bugs of whatever kind looks easier to achieve than adapting existing code for new functionality.
I might be wrong in 20 years, let's see.
Don't like ads? Remove them while supporting the forum:
Subscribe to Fastlane Insiders.
