Not a bot (although I have been accused of it, due to my activity here, and on GitHub, but I’ve been this way for longer than LLMs have been a thing. I’m retired, “on the spectrum,” and don’t participate in any other social media).
I’m currently working on a rewrite of an app that originally took two years. It’s been about three months, and I’m probably about 70% done. It’s a total “from scratch” rewrite; both client and server (two versions of each, as I also have administrative code). It’s a pretty big system, for one guy. I couldn’t do it, without the LLM.
It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
But I’m also not doing it in a manner that eats tokens. I just use the standard $20/month subscription as a chat. I suspect my workflow is not one that Anthropic or OpenAI really wants out there.
But I also bet that many HN accounts are bots; although I think many may be ones run by enthusiasts, not some AI cabal.
For 5 million comments like yours I haven't seen a single one with the old code vs. the new code. I understand that not all code is public that way, of course, and I don't mean to put you on the spot personally. But where are all the open source projects that now do the same with better error handling using less resources? Where are 100+ MB Electron apps reduced to more correct sizes like a few MB, or even a few dozen kB? Why aren't startup times getting slashed across the board? Why isn't RAM usage falling faster than RAM prices are increasing?
It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
But your anecdote is much more balanced and more in line with my personal findings. Not like all the AI astroturfing that happens here. I like to use LLMs as well, but in a very targeted way in places where LLMs shine.
Yes, LLMs can make you much more productive. But so could assembly -> C -> Python or Rust, or switching an IDE with code completion and support for refactoring. Each step makes you more productive.
Sure, LLMs can spit out greenfields projects. But on large projects with complex requirements, you still need senior engineers to guide them, carefully review the output, and as you say throw out code and write it from scratch in a better way.
I had some friends who ended up in bits of AI psychosis. They exclaim that a swarm of agents was writing all their code, but every time I ask them to show the end-result, all they have is a pile of code that they don't understand, nor doesn't really work either. At the same time, they stopped getting any actual work done.
At any rate, somebody had a great analogy on HN recently: think of it is a vector, LLMs can significantly increase the magnitude of the vector, but you still have to make sure that the orientation of the vector is correct.
> It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
Same here :)
> not some AI cabal.
There are enough enthusiasts to make it feel like one. Also an unhealthy doze of marketers, people buying into hype, AI psychosis etc.
Showing actual improved products and features. Showing actual code. etc.
Not a bot (although I have been accused of it, due to my activity here, and on GitHub, but I’ve been this way for longer than LLMs have been a thing. I’m retired, “on the spectrum,” and don’t participate in any other social media).
I’m currently working on a rewrite of an app that originally took two years. It’s been about three months, and I’m probably about 70% done. It’s a total “from scratch” rewrite; both client and server (two versions of each, as I also have administrative code). It’s a pretty big system, for one guy. I couldn’t do it, without the LLM.
It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
But I’m also not doing it in a manner that eats tokens. I just use the standard $20/month subscription as a chat. I suspect my workflow is not one that Anthropic or OpenAI really wants out there.
But I also bet that many HN accounts are bots; although I think many may be ones run by enthusiasts, not some AI cabal.
For 5 million comments like yours I haven't seen a single one with the old code vs. the new code. I understand that not all code is public that way, of course, and I don't mean to put you on the spot personally. But where are all the open source projects that now do the same with better error handling using less resources? Where are 100+ MB Electron apps reduced to more correct sizes like a few MB, or even a few dozen kB? Why aren't startup times getting slashed across the board? Why isn't RAM usage falling faster than RAM prices are increasing?
8 replies →
It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
But your anecdote is much more balanced and more in line with my personal findings. Not like all the AI astroturfing that happens here. I like to use LLMs as well, but in a very targeted way in places where LLMs shine.
Yes, LLMs can make you much more productive. But so could assembly -> C -> Python or Rust, or switching an IDE with code completion and support for refactoring. Each step makes you more productive.
Sure, LLMs can spit out greenfields projects. But on large projects with complex requirements, you still need senior engineers to guide them, carefully review the output, and as you say throw out code and write it from scratch in a better way.
I had some friends who ended up in bits of AI psychosis. They exclaim that a swarm of agents was writing all their code, but every time I ask them to show the end-result, all they have is a pile of code that they don't understand, nor doesn't really work either. At the same time, they stopped getting any actual work done.
At any rate, somebody had a great analogy on HN recently: think of it is a vector, LLMs can significantly increase the magnitude of the vector, but you still have to make sure that the orientation of the vector is correct.
1 reply →
> It’s not been a cakewalk. I’ve needed to toss out large swaths of LLM-generated code, and rewrite by hand, but, for the most part, it’s been a huge help.
Same here :)
> not some AI cabal.
There are enough enthusiasts to make it feel like one. Also an unhealthy doze of marketers, people buying into hype, AI psychosis etc.
1 reply →
I shipped a project at work in 3 months instead of the estimated original 6-9 months.
Sure you did. We can't see the project, of course, because she lives in Canada...
4 replies →