Comment by nicoburns
13 days ago
> Holy hell HN, downvoted to -4 in record time. Y'all don't like what's happening, but it's really happening. > > I'm not lying about this. > > I provided my background so you'd understand the context of my claims. I have a solid background in tech.
There are lots of people claiming this. Many of whom have a solid background. Every now and then I check out someone's claim (checking the code they've generated). I've yet to find an AI-generated codebase that passed that check so far.
Perhaps yours is the one that does, but as we can't see the code for ourselves, there's no way for us to really know. And it's hard to take your word for it when there are so many people falsely making the same claims.
I expect a lot of HNers have had this experience.
Have you looked at antirez's code?
https://github.com/antirez/flux2.c
Generating code from scratch and modifying existing code are two different things, obviously the latter being where AI doesn't do great. Carefully managing and compressing context can somewhat help, but that is far from being a perfect solution.
Every time someone asks for good examples of AI written code, and they are given one, they either move the goalposts or simply don't reply.
I went from using copilot autocomplete and mostly just disabling it because it was not great, to now using claude most of my day. I gets stuff wrong. It does generate inefficient code. It definitely has a way to go, but otherwise it just saves so much time.
1 reply →
But you can iterate over the code until you like it. If you don't review the code, you risk slop. But reviewing takes time, and most people don't do that in the prototyping stage, because the code might still significantly change. But as your project matures you become more critical of the code and you review it more closely. And over time you also get more feeling of how to work with the AI to get the results you want.