The problem is that there’s always a communication cost to outsourcing something. And you can’t outsource a whole project to an LLM, like a human intern. You’re just outsourcing one micro-, sub-task after another. To use the human analogy, it’s more like you’re standing over their shoulder and telling them what function to write, one after the other. And they’re SUPER fast with small, pure functions, but they get confused with anything else.
Is that a faster way to program?
Maybe? If you can fluidly decompose things into small functions in your head and the problem can be solved that way?
I don’t know though, I find myself using chatGPT for bigger meta questions much more frequently than the Copilot autocomplete
The problem is that there’s always a communication cost to outsourcing something. And you can’t outsource a whole project to an LLM, like a human intern. You’re just outsourcing one micro-, sub-task after another. To use the human analogy, it’s more like you’re standing over their shoulder and telling them what function to write, one after the other. And they’re SUPER fast with small, pure functions, but they get confused with anything else.
Is that a faster way to program?
Maybe? If you can fluidly decompose things into small functions in your head and the problem can be solved that way?
I don’t know though, I find myself using chatGPT for bigger meta questions much more frequently than the Copilot autocomplete