I just used Claude yesterday to add some functionality to an existing python script that interacts with AWS. It created an unnecessary loop where 2 of 3 iterations were effectively no-ops. So while it did ultimately provide what I needed, I still had to refactor what it generated in order to remove the useless loop.
Remember, it was trained on Stack Overflow… a self-selected collection of “beginner” questions with a lot of “this is good enough to get you unstuck” answers. If you want better, you have to specify and test to ensure you get better.
Oh, about the un-necessary loops thing. I was once hired by a PhD to build a software development department to convert his “academic matlab code” into an actual commercial application. One of my programmers was another PhD, and he successfully translated the matlab into C++ and it did run faster, and he parallelized it successfully (the latest hotness in 2005) but was still the slowest part of our whole operation. So, we did an old-fashioned code review and found an un-necessary loop, unwound that, got the exact same answers 100x faster. Apparently the PhDs were loop-slop blind, one had missed it for years and the other had been working on the translation for months without finding it.
I just used Claude yesterday to add some functionality to an existing python script that interacts with AWS. It created an unnecessary loop where 2 of 3 iterations were effectively no-ops. So while it did ultimately provide what I needed, I still had to refactor what it generated in order to remove the useless loop.
Remember, it was trained on Stack Overflow… a self-selected collection of “beginner” questions with a lot of “this is good enough to get you unstuck” answers. If you want better, you have to specify and test to ensure you get better.
Oh, about the un-necessary loops thing. I was once hired by a PhD to build a software development department to convert his “academic matlab code” into an actual commercial application. One of my programmers was another PhD, and he successfully translated the matlab into C++ and it did run faster, and he parallelized it successfully (the latest hotness in 2005) but was still the slowest part of our whole operation. So, we did an old-fashioned code review and found an un-necessary loop, unwound that, got the exact same answers 100x faster. Apparently the PhDs were loop-slop blind, one had missed it for years and the other had been working on the translation for months without finding it.