>Does anyone else find that LLM code generators improve your overall productivity marginally
at first it was marginal. Now it's far more than marginally, though like any tool it takes time to become proficient at it, and so you make lots of mistakes at first.
Those mistakes are probably an annoyance.
>Basically the extra cognitive annoyance you have to experience every time it generates the wrong thing
Do you find that you often code the same way? doing the wrong thing sometimes? Everyone makes mistakes.
>It’s like you can achieve the end result faster but it’s more mentally draining and annoying to keep tweaking the LLM than to just write it yourself
I think what you need to do is design your own benchmarks. Put various LLM through your benchmarks and judge, but also notice how each of the models tend to fail at doing something in your benchmark. so then you have to add it to the original benchmark prompt.
This will show you why and how you need to be more specific.
>It’s like you can achieve the end result faster but it’s more mentally draining and annoying to keep tweaking the LLM than to just write it yourself
The annonyance will always be there, sometimes you just want the llm to fix something, you agree with the fix, but it was a bad decision and need major refactor later.
