←back to thread

421 points briankelly | 1 comments | | HN request time: 0.243s | source
Show context
necovek ◴[] No.43575664[source]
The premise might possibly be true, but as an actually seasoned Python developer, I've taken a look at one file: https://github.com/dx-tooling/platform-problem-monitoring-co...

All of it smells of a (lousy) junior software engineer: from configuring root logger at the top, module level (which relies on module import caching not to be reapplied), over not using a stdlib config file parser and building one themselves, to a raciness in load_json where it's checked for file existence with an if and then carrying on as if the file is certainly there...

In a nutshell, if the rest of it is like this, it simply sucks.

replies(23): >>43575714 #>>43575764 #>>43575953 #>>43576545 #>>43576732 #>>43576977 #>>43577008 #>>43577017 #>>43577193 #>>43577214 #>>43577226 #>>43577314 #>>43577850 #>>43578934 #>>43578952 #>>43578973 #>>43579760 #>>43581498 #>>43582065 #>>43583922 #>>43585046 #>>43585094 #>>43587376 #
rybosome ◴[] No.43575714[source]
Ok - not wrong at all. Now take that feedback and put it in a prompt back to the LLM.

They’re very good at honing bad code into good code with good feedback. And when you can describe good code faster than you can write it - for instance it uses a library you’re not intimately familiar with - this kind of coding can be enormously productive.

replies(5): >>43575812 #>>43575838 #>>43575956 #>>43577317 #>>43578501 #
1. barrell ◴[] No.43578501[source]
I would not say it is “very good” at that. Maybe it’s “capable,” but my (ample) experience has been the opposite. I have found the more exact I describe a solution, the less likely it is to succeed. And the more of a solution it has come up with, the less likely it is to change its mind about things.

Every since ~4o models, there seems to be a pretty decent chance that you ask it to change something specific and it says it will and it spits out line for line identical code to what you just asked it to change.

I have had some really cool success with AI finding optimizations in my code, but only when specifically asked, and even then I just read the response as theory and go write it myself, often in 1-15% the LoC as the LLM