alt Hacker News> Do we really believe people will start becoming more ambitious with AI or will most suffer from skill atrophy and less agency?
But which skill is atrophying? As a programmer I'm really bad at converting human readable code into machine code because we have compilers to do that for us. I can't remember the last time I had to run "ld" by hand. That skill totally atrophied. But at the same time, AI has made me more ambitious. I'm trying projects I wouldn't have before and even completing some of them! I can't talk for "people", broadly, but I believe most people want to be their best and do good and do things.
Replies
10xDev • yesterday at 7:05 PM
Those with high agency may have even greater agency, but I can also see the inverse effect.
I can run the same compiler (or assembler) with the same options on the same source code 100 times, and I will get bit-for-bit identical output 100 times (well, aside from the compiler/linker inserting time/date stamps into metadata). Most people will not need to care about judging the output of the compiler. Only rarely will the compiler or assembler do something incorrect, which will require someone with specialized skills to debug and fix.
If I give the same LLM the same prompt 100 times, I will get 100 different programs written. Some of them will not work at all. Some of them will work, but will have major bugs or performance issues. Some of them will work well, but have subtle issues or edge cases that aren't handled properly. A few of them will work perfectly, or at least adequately enough for the task at hand.
Every single time you give the LLM instructions to do something, you need someone qualified and capable of reviewing the output to make sure it works properly. And while I would say you need someone reviewing the source code, even if you're just vibe-coding, you still need someone to test the program and make sure it works, and even that requires some specialized skill.
Maybe LLMs (or their next-gen replacements) will eventually become good enough that you'll get the same output every time for those 100 prompts, or at least close enough and functional enough for it not to matter. But we're not there yet, and I think that's a big huge "maybe". In the meantime, skill atrophy among programmers is a real, reported phenomenon with the current crop of LLMs. That is worrying.